Details of the Researcher

PHOTO

Takayuki Okatani
Section
Graduate School of Information Sciences
Job title
Professor
Degree
  • 博士(工学)(東京大学)

  • 修士(工学)(東京大学)

Research History 2

  • 2016/10 - Present
    理化学研究所革新知能統合研究センター

  • 2003/11 - 2004/10
    Center of Advanced Information Processing, Rutgers University

Education 2

  • The University of Tokyo Graduate School, Division of Engineering 計数工学

    - 1999/03

  • The University of Tokyo Faculty of Engineering 計数工

    - 1994/03

Professional Memberships 4

  • 計測自動制御学会

  • 電子情報通信学会

  • 情報処理学会

  • IEEE

Research Interests 6

  • multimodal AI

  • 多視点幾何

  • 画像理解

  • 画像認識

  • 深層学習

  • コンピュータビジョン

Research Areas 2

  • Informatics / Intelligent robotics / Computer Vision

  • Informatics / Perceptual information processing / Computer Vision

Awards 11

  1. 画像の認識・理解シンポジウム(MIRU)2017優秀論文賞

    2017/08 画像の認識・理解シンポジウム実行委員会

  2. Best paper honorable mention

    2016/10/30 PACLIC2016 award committee Recognizing Open-Vocabulary Relations between Objects in Images

  3. MIRU優秀発表賞

    2016/08/03 MIRU実行委員会 Learning to Describe Product Images from Noisy Online Data

  4. MIRU優秀発表賞

    2015/08/01 MIRU運営委員会 Joint Clothing Detection

  5. MIRUフロンティア賞

    2015/08/01 MIRU運営委員会 Scene Change Detection from a Pair of Roughly Registered Images Based on CNN Features and Superpixel Segmentation

  6. Sang Uk Lee Best Student Paper Award

    2014/12/01 ACCV award committee Separation of Reflection Components by Sparse Non-negative Matrix Factorization

  7. Best Application Paper Honorable Mention

    2014/12/01 ACCV award committee Massive City-scale Surface Condition Analysis using Ground and Aerial Imagery

  8. 画像の認識理解シンポジウム(MIRU2010) MIRU長尾賞

    2010/07 画像の認識・理解シンポジウム実行委員会

  9. 画像の認識理解シンポジウム(MIRU2007)優秀論文賞

    2007/07/31 電子情報通信学会,情報処理学会他

  10. 画像の認識理解シンポジウム(MIRU2006)優秀論文賞

    2006/07/20 電子情報通信学会,情報処理学会他

  11. 情報処理学会平成9年度山下記念研究賞

    1997/01/01 情報処理学会

Show all ︎Show 5

Papers 175

  1. Inverting the Generation Process of Denoising Diffusion Implicit Models: Empirical Evaluation and a Novel Method.

    Yan Zeng, Masanori Suganuma, Takayuki Okatani

    WACV 4516-4524 2025

    DOI: 10.1109/WACV61041.2025.00453  

  2. RefVSR++: Exploiting Reference Inputs for Reference-based Video Super-resolution.

    Han Zou, Masanori Suganuma, Takayuki Okatani

    WACV 2756-2765 2025

    DOI: 10.1109/WACV61041.2025.00273  

  3. Deep learning-based bridge damage cause estimation from multiple images using visual question answering

    Tatsuro Yamane, Pang-jo Chun, Ji Dang, Takayuki Okatani

    Structure and Infrastructure Engineering 1-14 2024/05/20

    Publisher: Informa UK Limited

    DOI: 10.1080/15732479.2024.2355929  

    ISSN: 1573-2479

    eISSN: 1744-8980

  4. Symmetry-aware Neural Architecture for Embodied Visual Navigation.

    Shuang Liu, Masanori Suganuma, Takayuki Okatani

    International Journal of Computer Vision 132 (4) 1091-1107 2024/04

    DOI: 10.1007/s11263-023-01909-4  

  5. That's BAD: blind anomaly detection by implicit local feature clustering.

    Jie Zhang, Masanori Suganuma, Takayuki Okatani

    Machine Vision and Applications 35 (2) 31-31 2024/03

    DOI: 10.1007/s00138-024-01511-9  

  6. Improving visual question answering for bridge inspection by pre-training with external data of image-text pairs.

    Thannarot Kunlamai, Tatsuro Yamane, Masanori Suganuma, Pang-jo Chun, Takayuki Okatani

    Computer-Aided Civil and Infrastructure Engineering 39 (3) 345-361 2024/02

    DOI: 10.1111/mice.13086  

  7. Action-Agnostic Point-Level Supervision for Temporal Action Detection.

    Shuhei M. Yoshida, Takashi Shibata 0001, Makoto Terao, Takayuki Okatani, Masashi Sugiyama

    CoRR abs/2412.21205 2024

    DOI: 10.48550/arXiv.2412.21205  

  8. Exploring the Potential of Multi-Modal AI for Driving Hazard Prediction

    Korawat Charoenpitaks, Van Quang Nguyen, Masanori Suganuma, Masahiro Takahashi, Ryoma Niihara, Takayuki Okatani

    IEEE Transactions on Intelligent Vehicles 2024

    DOI: 10.1109/TIV.2024.3417353  

    eISSN: 2379-8858

  9. Appearance-Based Curriculum for Semi-Supervised Learning with Multi-Angle Unlabeled Data.

    Yuki Tanaka, Shuhei M. Yoshida, Takashi Shibata, Makoto Terao, Takayuki Okatani, Masashi Sugiyama

    IEEE/CVF Winter Conference on Applications of Computer Vision(WACV) 2768-2777 2024

    Publisher: IEEE

    DOI: 10.1109/WACV57701.2024.00276  

  10. SBCFormer: Lightweight Network Capable of Full-size ImageNet Classification at 1 FPS on Single Board Computers.

    Xiangyong Lu, Masanori Suganuma, Takayuki Okatani

    IEEE/CVF Winter Conference on Applications of Computer Vision(WACV) 1112-1122 2024

    Publisher: IEEE

    DOI: 10.1109/WACV57701.2024.00116  

  11. Contextual Affinity Distillation for Image Anomaly Detection.

    Jie Zhang, Masanori Suganuma, Takayuki Okatani

    IEEE/CVF Winter Conference on Applications of Computer Vision(WACV) 148-157 2024

    Publisher: IEEE

    DOI: 10.1109/WACV57701.2024.00022  

  12. Globalizing Local Features: Image Retrieval Using Shared Local Features with Pose Estimation for Faster Visual Localization.

    Wenzheng Song, Ran Yan, Boshu Lei, Takayuki Okatani

    IEEE International Conference on Robotics and Automation(ICRA) 6290-6297 2024

    Publisher: IEEE

    DOI: 10.1109/ICRA57147.2024.10610786  

  13. Temporal Insight Enhancement: Mitigating Temporal Hallucination in Video Understanding by Multimodal Large Language Models.

    Li Sun 0007, Liuan Wang, Jun Sun 0004, Takayuki Okatani

    ICPR (7) 455-473 2024

    DOI: 10.1007/978-3-031-78183-4_29  

  14. Rethinking unsupervised domain adaptation for semantic segmentation.

    Zhijie Wang, Masanori Suganuma, Takayuki Okatani

    Pattern Recognition Letters 186 119-125 2024

    DOI: 10.1016/j.patrec.2024.09.022  

  15. Improved high dynamic range imaging using multi-scale feature flows balanced between task-orientedness and accuracy.

    Qian Ye, Masanori Suganuma, Takayuki Okatani

    Computer Vision and Image Understanding 248 104126-104126 2024

    DOI: 10.1016/j.cviu.2024.104126  

  16. Unsupervised domain adaptation for semantic segmentation via cross-region alignment

    Zhijie Wang, Xing Liu, Masanori Suganuma, Takayuki Okatani

    Computer Vision and Image Understanding 234 2023/09

    DOI: 10.1016/j.cviu.2023.103743  

    ISSN: 1077-3142

    eISSN: 1090-235X

  17. Zero-shot temporal event localisation: Label-free, training-free, domain-free

    Li Sun, Ping Wang, Liuan Wang, Jun Sun, Takayuki Okatani

    IET Computer Vision 17 (5) 599-613 2023/08

    DOI: 10.1049/cvi2.12224  

    ISSN: 1751-9632

    eISSN: 1751-9640

  18. B-CNN: a deep learning method for accelerometer-based fatigue cracks monitoring system

    Yanjie Zhu, Hidehiko Sekiya, Takayuki Okatani, Masayuki Tai, Shogo Morichika

    Journal of Civil Structural Health Monitoring 13 (4-5) 947-959 2023/03/31

    Publisher: Springer Science and Business Media LLC

    DOI: 10.1007/s13349-023-00690-9  

    ISSN: 2190-5452

    eISSN: 2190-5479

  19. Zero-shot versus Many-shot: Unsupervised Texture Anomaly Detection

    Toshimichi Aota, Lloyd Teh Tzer Tong, Takayuki Okatani

    Proceedings - 2023 IEEE Winter Conference on Applications of Computer Vision, WACV 2023 5553-5561 2023

    DOI: 10.1109/WACV56688.2023.00552  

  20. Bright as the Sun: In-depth Analysis of Imagination-Driven Image Captioning

    Huyen Thi Thanh Tran, Takayuki Okatani

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 13844 LNCS 675-691 2023

    DOI: 10.1007/978-3-031-26316-3_40  

    ISSN: 0302-9743

    eISSN: 1611-3349

  21. Improving visual question answering for bridge inspection by pre-training with external data of image–text pairs

    Thannarot Kunlamai, Tatsuro Yamane, Masanori Suganuma, Pang Jo Chun, Takayaki Okatani

    Computer-Aided Civil and Infrastructure Engineering 2023

    DOI: 10.1111/mice.13086  

    ISSN: 1093-9687

    eISSN: 1467-8667

  22. How Do Label Errors Affect Thin Crack Detection by DNNs

    Liang Xu, Han Zou, Takayuki Okatani

    IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops 2023-June 4414-4423 2023

    DOI: 10.1109/CVPRW59228.2023.00464  

    ISSN: 2160-7508

    eISSN: 2160-7516

  23. Geometry Enhanced Reference-based Image Super-resolution

    Han Zou, Liang Xu, Takayuki Okatani

    IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops 2023-June 6124-6133 2023

    DOI: 10.1109/CVPRW59228.2023.00652  

    ISSN: 2160-7508

    eISSN: 2160-7516

  24. Network Pruning and Fine-tuning for Few-shot Industrial Image Anomaly Detection

    Jie Zhang, Masanori Suganuma, Takayuki Okatani

    IEEE International Conference on Industrial Informatics (INDIN) 2023-July 2023

    DOI: 10.1109/INDIN51400.2023.10218283  

    ISSN: 1935-4576

  25. Prompt Prototype Learning Based on Ranking Instruction For Few-Shot Visual Tasks

    Li Sun, Liuan Wang, Jun Sun, Takayuki Okatani

    Proceedings - International Conference on Image Processing, ICIP 3235-3239 2023

    DOI: 10.1109/ICIP49359.2023.10222039  

    ISSN: 1522-4880

  26. Accurate Single-Image Defocus Deblurring Based on Improved Integration with Defocus Map Estimation

    Qian Ye, Masanori Suganuma, Takayuki Okatani

    Proceedings - International Conference on Image Processing, ICIP 750-754 2023

    DOI: 10.1109/ICIP49359.2023.10223146  

    ISSN: 1522-4880

  27. Bridge Damage Cause Estimation Using Multiple Images Based on Visual Question Answering.

    Tatsuro Yamane, Pang-jo Chun, Ji Dang, Takayuki Okatani

    CoRR abs/2302.09208 2023

    DOI: 10.48550/arXiv.2302.09208  

  28. A lightweight deep learning model for automatic segmentation and analysis of ophthalmic images

    Parmanand Sharma, Takahiro Ninomiya, Kazuko Omodaka, Naoki Takahashi, Takehiro Miya, Noriko Himori, Takayuki Okatani, Toru Nakazawa

    Scientific Reports 12 (1) 2022/12

    Publisher: Springer Science and Business Media LLC

    DOI: 10.1038/s41598-022-12486-w  

    eISSN: 2045-2322

    More details Close

    Abstract Detection, diagnosis, and treatment of ophthalmic diseases depend on extraction of information (features and/or their dimensions) from the images. Deep learning (DL) model are crucial for the automation of it. Here, we report on the development of a lightweight DL model, which can precisely segment/detect the required features automatically. The model utilizes dimensionality reduction of image to extract important features, and channel contraction to allow only the required high-level features necessary for reconstruction of segmented feature image. Performance of present model in detection of glaucoma from optical coherence tomography angiography (OCTA) images of retina is high (area under the receiver-operator characteristic curve AUC ~ 0.81). Bland–Altman analysis gave exceptionally low bias (~ 0.00185), and high Pearson’s correlation coefficient (p = 0.9969) between the parameters determined from manual and DL based segmentation. On the same dataset, bias is an order of magnitude higher (~ 0.0694, p = 0.8534) for commercial software. Present model is 10 times lighter than Unet (popular for biomedical image segmentation) and have a better segmentation accuracy and model training reproducibility (based on the analysis of 3670 OCTA images). High dice similarity coefficient (D) for variety of ophthalmic images suggested it’s wider scope in precise segmentation of images even from other fields. Our concept of channel narrowing is not only important for the segmentation problems, but it can also reduce number of parameters significantly in object classification models. Enhanced disease diagnostic accuracy can be achieved for the resource limited devices (such as mobile phone, Nvidia’s Jetson, Raspberry pi) used in self-monitoring, and tele-screening (memory size of trained model ~ 35 MB).

  29. Development of a continuum robot enhanced with distributed sensors for search and rescue

    Yu Yamauchi, Yuichi Ambe, Hikaru Nagano, Masashi Konyo, Yoshiaki Bando, Eisuke Ito, Solvi Arnold, Kimitoshi Yamazaki, Katsutoshi Itoyama, Takayuki Okatani, Hiroshi G. Okuno, Satoshi Tadokoro

    ROBOMECH Journal 9 (1) 2022/12

    Publisher: Springer Science and Business Media LLC

    DOI: 10.1186/s40648-022-00223-x  

    eISSN: 2197-4225

    More details Close

    Abstract Continuum robots can enter narrow spaces and are useful for search and rescue missions in disaster sites. The exploration efficiency at disaster sites improves if the robots can simultaneously acquire several pieces of information. However, a continuum robot that can simultaneously acquire information to such an extent has not yet been designed. This is because attaching multiple sensors to the robot without compromising its body flexibility is challenging. In this study, we installed multiple small sensors in a distributed manner to develop a continuum-robot system with multiple information-gathering functions. In addition, a field experiment with the robot demonstrated that the gathered multiple information has a potential to improve the searching efficiency. Concretely, we developed an active scope camera with sensory functions, which was equipped with a total of 80 distributed sensors, such as inertial measurement units, microphones, speakers, and vibration sensors. Herein, we consider space-saving, noise reduction, and the ease of maintenance for designing the robot. The developed robot can communicate with all the attached sensors even if it is bent with a minimum bending radius of 250 mm. We also developed an operation interface that integrates search-support technologies using the information gathered via sensors. We demonstrated the survivor search procedure in a simulated rubble environment of the Fukushima Robot Test Field. We confirmed that the information provided through the operation interface is useful for searching and finding survivors. The limitations of the designed system are also discussed. The development of such a continuum robot system, with a great potential for several applications, extends the application of continuum robots to disaster management and will benefit the community at large.

  30. GRIT: Faster and Better Image Captioning Transformer Using Dual Visual Features.

    Van-Quang Nguyen, Masanori Suganuma, Takayuki Okatani

    European Conference on Computer Vision (ECCV) 36 167-184 2022/09

    DOI: 10.1007/978-3-031-20059-5_10  

  31. Real-time vehicle identification using two-step LSTM method for acceleration-based bridge weigh-in-motion system

    Yanjie Zhu, Hidehiko Sekiya, Takayuki Okatani, Ikumasa Yoshida, Shuichi Hirano

    JOURNAL OF CIVIL STRUCTURAL HEALTH MONITORING 12 (3) 689-703 2022/06

    DOI: 10.1007/s13349-022-00576-2  

    ISSN: 2190-5452

    eISSN: 2190-5479

  32. Symmetry-aware Neural Architecture for Embodied Visual Exploration.

    Shuang Liu 0002, Takayuki Okatani

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 17221-17230 2022/06

    DOI: 10.1109/CVPR52688.2022.01673  

    ISSN: 0920-5691

    eISSN: 1573-1405

  33. AUTOMATIC DETECTION OF INNER DEFECTS OF CONCRETE BY ANALYZING THERMAL IMAGES USING SELF-SUPERVISED LEARNING

    KAWANOWA Sota, HAYASHI Shogo, OKATANI Takayuki, LIU Kang-Jun, CHUN Pang-jo

    Intelligence, Informatics and Infrastructure 3 (J2) 47-55 2022

    Publisher: Japan Society of Civil Engineers

    DOI: 10.11532/jsceiii.3.j2_47  

    eISSN: 2435-9262

    More details Close

    In the infrared thermography method, which remotely detects inner defects by capturing thermal images of concrete, damaged areas are often overlooked by human judgment. Although there is a movement to introduce CNN-based automatic detection to the infrared method, sufficient accuracy has not been obtained due to the lack of training data. Therefore, in this study, we focus on self-supervised learning. Self-supervised learning has the potential to achieve high accuracy with fewer teacher labels. In this study, we present an example of how to introduce self-supervised learning to the infrared thermography method and verify its effectiveness.

  34. Deep learning increases the availability of organism photographs taken by citizens in citizen science programs

    Yukari Suzuki-Ohno, Thomas Westfechtel, Jun Yokoyama, Kazunori Ohno, Tohru Nakashizuka, Masakado Kawata, Takayuki Okatani

    SCIENTIFIC REPORTS 12 (1) 2022/01

    DOI: 10.1038/s41598-022-05163-5  

    ISSN: 2045-2322

  35. Single-image Defocus Deblurring by Integration of Defocus Map Prediction Tracing the Inverse Problem Computation.

    Qian Ye, Masanori Suganuma, Takayuki Okatani

    CoRR abs/2207.03047 2022

    DOI: 10.48550/arXiv.2207.03047  

  36. Learning Regularized Multi-Scale Feature Flow for High Dynamic Range Imaging.

    Qian Ye, Masanori Suganuma, Jun Xiao, Takayuki Okatani

    CoRR abs/2207.02539 2022

    DOI: 10.48550/arXiv.2207.02539  

  37. Rethinking Unsupervised Domain Adaptation for Semantic Segmentation.

    Zhijie Wang, Masanori Suganuma, Takayuki Okatani

    CoRR abs/2207.00067 2022

    DOI: 10.48550/arXiv.2207.00067  

  38. Learning to Bundle-adjust: A Graph Network Approach to Faster Optimization of Bundle Adjustment for Vehicular SLAM

    Tetsuya Tanaka, Yukihiro Sasagawa, Takayuki Okatani

    2021 IEEE/CVF International Conference on Computer Vision (ICCV) 6230-6239 2021/10

    Publisher: IEEE

    DOI: 10.1109/iccv48922.2021.00619  

  39. Acceleration-Based Deep Learning Method for Vehicle Monitoring

    Yanjie Zhu, Hidehiko Sekiya, Takayuki Okatani, Ikumasa Yoshida, Shuichi Hirano

    IEEE SENSORS JOURNAL 21 (15) 17154-17161 2021/08

    DOI: 10.1109/JSEN.2021.3082145  

    ISSN: 1530-437X

    eISSN: 1558-1748

  40. EFFECTS OF CNN MODELS AND DIFFIRENT DIGITAL IMAGE RESOLUTION ON ACCURACY OF RUST CONDITION RATING OF WEATHERING STEEL

    TAI Masayuki, SEKIYA Hidehiko, OKATANI Takayuki, NAKAMURA Shozo, SHIMIZU Takashi

    Intelligence, Informatics and Infrastructure 2 (J2) 378-385 2021

    Publisher: Japan Society of Civil Engineers

    DOI: 10.11532/jsceiii.2.j2_378  

    eISSN: 2435-9262

    More details Close

    The inspection and diagnosis of weathering steel bridges is based on the deterioration of anti-corrosion function, which is evaluated with the corrosion condition rating by observation. However, in order to perform an accurate evaluation by therating, inspectors need to have necessary to have adequate experience. Due to the recent shortage of human resources and inspection costs, it is needed to establish a simple and accurate evaluation method. In this study, the identification of rust condition has been investigated by using digital images of surface rust condition in weathering steel bridges and existing CNN models. Inaddition, the effect of digital image resolution on the identification accuracy has been considered. As a result, the accuracy isrelatively high in the case of VGG19 and SEnet. The larger input image size gives the better accuracy. Furthermore, the accuracy tends to decrease when the resolution of the images used for training and validation is different.

  41. Progressive and Selective Fusion Network for High Dynamic Range Imaging.

    Qian Ye, Jun Xiao, Kin-Man Lam 0001, Takayuki Okatani

    MM '21: ACM Multimedia Conference 5290-5297 2021

    Publisher: ACM

    DOI: 10.1145/3474085.3475651  

  42. Look Wide and Interpret Twice: Improving Performance on Interactive Instruction-following Tasks.

    Van-Quang Nguyen, Masanori Suganuma, Takayuki Okatani

    Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence(IJCAI) 923-930 2021

    Publisher: ijcai.org

    DOI: 10.24963/ijcai.2021/128  

  43. Matching in the Dark: A Dataset for Matching Image Pairs of Low-light Scenes.

    Wenzheng Song, Masanori Suganuma, Xing Liu 0010, Noriyuki Shimobayashi, Daisuke Maruta, Takayuki Okatani

    2021 IEEE/CVF International Conference on Computer Vision(ICCV) 6009-6018 2021

    Publisher: IEEE

    DOI: 10.1109/ICCV48922.2021.00597  

  44. Symbolizing Visual Features for Pre-training with Unlabeled Images.

    Yuichi Kamata, Moyuru Yamada, Keizo Kato, Akira Nakagawa, Takayuki Okatani

    Pattern Recognition - 6th Asian Conference 490-503 2021

    Publisher: Springer

    DOI: 10.1007/978-3-031-02444-3_37  

  45. Innovative technologies for infrastructure construction and maintenance through collaborative robots based on an open design approach.

    Keiji Nagatani, Masato Abe, Koichi Osuka, Pang-jo Chun, Takayuki Okatani, Mayuko Nishio, Shota Chikushi, Takamitsu Matsubara, Yusuke Ikemoto, Hajime Asama

    Advanced Robotics 35 (11) 715-722 2021

    DOI: 10.1080/01691864.2021.1929471  

  46. Deep Learning for Channel-Agnostic Brain Decoding across Multiple Subjects

    Hiroto Date, Keisuke Kawasaki, Isao Hasegawa, Takayuki Okatani

    8th International Winter Conference on Brain-Computer Interface, BCI 2020 2020/02/01

    Publisher: Institute of Electrical and Electronics Engineers Inc.

    DOI: 10.1109/BCI48061.2020.9061615  

  47. Toward Explainable Fashion Recommendation. Peer-reviewed

    Pongsate Tangseng, Takayuki Okatani

    IEEE Winter Conference on Applications of Computer Vision(WACV) 2142-2151 2020

    Publisher: IEEE

    DOI: 10.1109/WACV45572.2020.9093367  

  48. Analysis and a Solution of Momentarily Missed Detection for Anchor-based Object Detectors.

    Yusuke Hosoya, Masanori Suganuma, Takayuki Okatani

    IEEE Winter Conference on Applications of Computer Vision(WACV) 1399-1407 2020

    Publisher: IEEE

    DOI: 10.1109/WACV45572.2020.9093553  

  49. Efficient Attention Mechanism for Visual Dialog that Can Handle All the Interactions Between Multiple Inputs.

    Van-Quang Nguyen, Masanori Suganuma, Takayuki Okatani

    Computer Vision - ECCV 2020 - 16th European Conference 223-240 2020

    Publisher: Springer

    DOI: 10.1007/978-3-030-58586-0_14  

  50. Removal of Image Obstacles for Vehicle-mounted Surrounding Monitoring Cameras by Real-time Video Inpainting.

    Yoshihiro Hirohashi, Kenichi Narioka, Masanori Suganuma, Xing Liu 0010, Yukimasa Tamatsu, Takayuki Okatani

    2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition 857-866 2020

    Publisher: Computer Vision Foundation / IEEE

    DOI: 10.1109/CVPRW50498.2020.00115  

  51. Hyperparameter-Free Out-of-Distribution Detection Using Cosine Similarity.

    Engkarat Techapanurak, Masanori Suganuma, Takayuki Okatani

    Computer Vision - ACCV 2020 - 15th Asian Conference on Computer Vision 53-69 2020

    Publisher: Springer

    DOI: 10.1007/978-3-030-69538-5_4  

  52. Extending information maximization from a rate-distortion perspective.

    Yan Zhang 0055, Junjie Hu, Takayuki Okatani

    Neurocomputing 399 285-295 2020

    DOI: 10.1016/j.neucom.2020.02.061  

    ISSN: 0925-2312

    eISSN: 1872-8286

  53. A vision-based method for crack detection in gusset plate welded joints of steel bridges using deep convolutional neural networks

    Cao Vu Dung, Hidehiko Sekiya, Suichi Hirano, Takayuki Okatani, Chitoshi Miki

    Automation in Construction 102 217-229 2019/06/01

    DOI: 10.1016/j.autcon.2019.02.013  

    ISSN: 0926-5805

  54. Attention-based Adaptive Selection of Operations for Image Restoration in the Presence of Unknown Combined Distortions. Peer-reviewed

    Masanori Suganuma, Xing Liu, Takayuki Okatani

    Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019/06

  55. Multi-task Learning of Hierarchical Vision-Language Representation. Peer-reviewed

    Duy-Kien Nguyen, Takayuki Okatani

    Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019/06

  56. Dual Residual Networks Leveraging the Potential of Paired Operations for Image Restoration. Peer-reviewed

    Xing Liu, Masanori Suganuma, Zhun Sun, Takayuki Okatani

    Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2019/06

  57. A Generative Model of Underwater Images for Active Landmark Detection and Docking.

    Shuang Liu 0002, Mete Özay, Hongli Xu, Yang Lin, Takayuki Okatani

    2019 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS) 8034-8039 2019

    Publisher: IEEE

    DOI: 10.1109/IROS40897.2019.8968146  

  58. Visualization of Convolutional Neural Networks for Monocular Depth Estimation.

    Junjie Hu, Yan Zhang 0055, Takayuki Okatani

    2019 IEEE/CVF International Conference on Computer Vision(ICCV) 3868-3877 2019

    Publisher: IEEE

    DOI: 10.1109/ICCV.2019.00397  

  59. Improving Head Pose Estimation with a Combined Loss and Bounding Box Margin Adjustment.

    Mingzhen Shao, Zhun Sun, Mete Ozay, Takayuki Okatani

    14th IEEE International Conference on Automatic Face & Gesture Recognition(FG) 1-5 2019

    Publisher: IEEE

    DOI: 10.1109/FG.2019.8756605  

  60. Deep Learning for Natural Image Reconstruction from Electrocorticography Signals.

    Hiroto Date, Keisuke Kawasaki, Isao Hasegawa, Takayuki Okatani

    2019 IEEE International Conference on Bioinformatics and Biomedicine(BIBM) 2331-2336 2019

    Publisher: IEEE

    DOI: 10.1109/BIBM47256.2019.8983029  

  61. ImPACT-TRC Thin Serpentine Robot Platform for Urban Search and Rescue. Peer-reviewed

    Masashi Konyo, Yuichi Ambe, Hikaru Nagano, Yu Yamauchi, Satoshi Tadokoro, Yoshiaki Bando, Katsutoshi Itoyama, Hiroshi G. Okuno, Takayuki Okatani, Kanta Shimizu, Eisuke Ito

    Disaster Robotics - Results from the ImPACT Tough Robotics Challenge 25-76 2019

    Publisher: Springer

    DOI: 10.1007/978-3-030-05321-5_2  

  62. Video-Rate Video Inpainting. Peer-reviewed

    Rito Murase, Yan Zhang, Takayuki Okatani

    IEEE Winter Conference on Applications of Computer Vision, WACV 2019, Waikoloa Village, HI, USA, January 7-11, 2019 1553-1561 2019

    Publisher: IEEE

    DOI: 10.1109/WACV.2019.00170  

  63. Revisiting Single Image Depth Estimation: Toward Higher Resolution Maps With Accurate Object Boundaries. Peer-reviewed

    Junjie Hu, Mete Ozay, Yan Zhang, Takayuki Okatani

    IEEE Winter Conference on Applications of Computer Vision, WACV 2019, Waikoloa Village, HI, USA, January 7-11, 2019 1043-1051 2019

    Publisher: IEEE

    DOI: 10.1109/WACV.2019.00116  

  64. Improving Generalization Ability of Deep Neural Networks for Visual Recognition Tasks. Peer-reviewed

    Takayuki Okatani, Xing Liu 0010, Masanori Suganuma

    Computational Color Imaging - 7th International Workshop, CCIW 2019, Chiba, Japan, March 27-29, 2019, Proceedings 3-13 2019

    Publisher: Springer

    DOI: 10.1007/978-3-030-13940-7_1  

  65. Detection and Pose Estimation for Short-Range Vision-Based Underwater Docking. Peer-reviewed

    Shuang Liu 0002, Mete Ozay, Takayuki Okatani, Hongli Xu, Kai Sun, Yang Lin

    IEEE Access 7 2720-2749 2019

    DOI: 10.1109/ACCESS.2018.2885537  

  66. Learning deep representations and detection of docking stations using underwater imaging

    Shuang Liu, Mete Ozay, Takayuki Okatani, Hongli Xu, Yang Lin, Haitao Gu

    2018 OCEANS - MTS/IEEE Kobe Techno-Oceans, OCEANS - Kobe 2018 2018/12/04

    Publisher: Institute of Electrical and Electronics Engineers Inc.

    DOI: 10.1109/OCEANSKOBE.2018.8559067  

  67. Recommending Outfits from Personal Closet Peer-reviewed

    Pongsate Tangseng, Kota Yamaguchi, Takayuki Okatani

    Proceedings - 2017 IEEE International Conference on Computer Vision Workshops, ICCVW 2017 2018- 2275-2279 2018/01/19

    Publisher: Institute of Electrical and Electronics Engineers Inc.

    DOI: 10.1109/ICCVW.2017.267  

  68. Perturbation Tolerance of Deep Neural Networks and Humans in Material Recognition Peer-reviewed

    Xing Liu, Masataka Sawayama, Ryusuke Hayashi, Mete Ozay, Takayuki Okatani, Shin'ya Nishida

    Journal of Vision 18 (10) 756-756 2018

    DOI: 10.1167/18.10.756  

  69. Deep Structured Energy-Based Image Inpainting. Peer-reviewed

    Fazil Altinel, Mete Ozay, Takayuki Okatani

    24th International Conference on Pattern Recognition, ICPR 2018, Beijing, China, August 20-24, 2018 423-428 2018

    Publisher: IEEE Computer Society

    DOI: 10.1109/ICPR.2018.8546025  

  70. Exploiting the Potential of Standard Convolutional Autoencoders for Image Restoration by Evolutionary Search. Peer-reviewed

    Masanori Suganuma, Mete Ozay, Takayuki Okatani

    Proceedings of the 35th International Conference on Machine Learning, ICML 2018, Stockholmsmässan, Stockholm, Sweden, July 10-15, 2018 4778-4787 2018

    Publisher: JMLR.org

  71. Feature Quantization for Defending Against Distortion of Images. Peer-reviewed

    Zhun Sun, Mete Ozay, Yan Zhang, Xing Liu, Takayuki Okatani

    2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018 7957-7966 2018

    Publisher: IEEE Computer Society

  72. Improved Fusion of Visual and Language Representations by Dense Symmetric Co-Attention for Visual Question Answering. Peer-reviewed

    Duy-Kien Nguyen, Takayuki Okatani

    2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018 6087-6096 2018

    Publisher: IEEE Computer Society

  73. Training CNNs With Normalized Kernels. Peer-reviewed

    Mete Özay, Takayuki Okatani

    Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence, (AAAI-18), the 30th innovative Applications of Artificial Intelligence (IAAI-18), and the 8th AAAI Symposium on Educational Advances in Artificial Intelligence (EAAI-18), New Orl 3884-3891 2018

    Publisher: AAAI Press

  74. Truncating Wide Networks Using Binary Tree Architectures Peer-reviewed

    Yan Zhangy, Mete Ozayy, Shuohao Li, Takayuki Okatani

    Proceedings of the IEEE International Conference on Computer Vision 2017- 2116-2124 2017/12/22

    DOI: 10.1109/ICCV.2017.231  

    ISSN: 1550-5499

  75. Attention to describe products with attributes Peer-reviewed

    Shuohao Li, Kota Yamaguchi, Takayuki Okatani

    Proceedings of the 15th IAPR International Conference on Machine Vision Applications, MVA 2017 215-218 2017/07/19

    Publisher: Institute of Electrical and Electronics Engineers Inc.

    DOI: 10.23919/MVA.2017.7986839  

  76. Temporal city modeling using street level imagery Peer-reviewed

    Ken Sakurada, Daiki Tetsuka, Takayuki Okatani

    COMPUTER VISION AND IMAGE UNDERSTANDING 157 55-71 2017/04

    DOI: 10.1016/j.cviu.2017.01.012  

    ISSN: 1077-3142

    eISSN: 1090-235X

  77. Learning to describe E-commerce images from noisy online data Peer-reviewed

    Takuya Yashima, Naoaki Okazaki, Kentaro Inui, Kota Yamaguchi, Takayuki Okatani

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 10115 85-100 2017

    DOI: 10.1007/978-3-319-54193-8_6  

    ISSN: 1611-3349 0302-9743

    eISSN: 1611-3349

  78. Self-calibration-based Approach to Critical Motion Sequences of Rolling-shutter Structure from Motion Peer-reviewed

    Eisuke Ito, Takayuki Okatani

    30TH IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2017) 4512-4520 2017

    DOI: 10.1109/CVPR.2017.480  

    ISSN: 1063-6919

  79. Self-calibration-based Approach to Critical Motion Sequences of Rolling-shutter Structure from Motion Peer-reviewed

    Eisuke Ito, Takayuki Okatani

    30TH IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2017) 4512-4520 2017

    DOI: 10.1109/CVPR.2017.480  

    ISSN: 1063-6919

  80. Separation of reflection components by sparse non-negative matrix factorization Peer-reviewed

    Yasushi Akashi, Takayuki Okatani

    COMPUTER VISION AND IMAGE UNDERSTANDING 146 77-85 2016/05

    DOI: 10.1016/j.cviu.2015.09.001  

    ISSN: 1077-3142

    eISSN: 1090-235X

  81. Hybrid macro-micro visual analysis for city-scale state estimation Peer-reviewed

    Ken Sakurada, Takayuki Okatani, Kris M. Kitani

    COMPUTER VISION AND IMAGE UNDERSTANDING 146 86-98 2016/05

    DOI: 10.1016/j.cviu.2016.02.017  

    ISSN: 1077-3142

    eISSN: 1090-235X

  82. A Gaze-Reactive Display for Simulating Depth-of-Field of Eyes When Viewing Scenes with Multiple Depths Peer-reviewed

    Tatsuro Orikasa, Takayuki Okatani

    IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS E99D (3) 739-746 2016/03

    DOI: 10.1587/transinf.2015EDP7110  

    ISSN: 1745-1361

  83. Design of Kernels in Convolutional Neural Networks for Image Classification Peer-reviewed

    Zhun Sun, Mete Ozay, Takayuki Okatani

    COMPUTER VISION - ECCV 2016, PT VII 9911 51-66 2016

    DOI: 10.1007/978-3-319-46478-7_4  

    ISSN: 0302-9743

  84. Automatic Attribute Discovery with Neural Activations Peer-reviewed

    Sirion Vittayakorn, Takayuki Umeda, Kazuhiko Murasaki, Kyoko Sudo, Takayuki Okatani, Kota Yamaguchi

    COMPUTER VISION - ECCV 2016, PT IV 9908 252-268 2016

    DOI: 10.1007/978-3-319-46493-0_16  

    ISSN: 0302-9743

  85. Recognizing open-vocabulary relations between objects in images Peer-reviewed

    Muraoka, M, Maharjan, S, Saito, M, Yamaguchi, K, Okazaki, N, Okatani, T, Inui, K

    Proceedings of the 30th Pacific Asia Conference on Language, Information and Computation, PACLIC 2016 239-248 2016

  86. Integrating Deep Features for Material Recognition Peer-reviewed

    Yan Zhang, Mete Ozay, Xing Liu, Takayuki Okatani

    2016 23RD INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR) 3697-3702 2016

    DOI: 10.1109/ICPR.2016.7900209  

    ISSN: 1051-4651

  87. Possibility to use product image and review text based on the association between onomatopoeia and texture Peer-reviewed

    Ryuichi Doizaki, Saki Iiba, Takayuki Okatani, Maki Sakamoto

    Transactions of the Japanese Society for Artificial Intelligence 30 (1) 124-137 2015/01/06

    Publisher: Japanese Society for Artificial Intelligence

    DOI: 10.1527/tjsai.30.124  

    ISSN: 1346-8030 1346-0714

  88. Change Detection from a Street Image Pair using CNN Features and Superpixel Segmentation.

    Ken Sakurada, Takayuki Okatani

    Proceedings of the British Machine Vision Conference 2015(BMVC) 61-12 2015

    Publisher: BMVA Press

    DOI: 10.5244/C.29.61  

  89. Transformation of Markov Random Fields for Marginal Distribution Estimation Peer-reviewed

    Masaki Saito, Takayuki Okatani

    2015 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) 07-12-June-2015 797-805 2015

    DOI: 10.1109/CVPR.2015.7298680  

    ISSN: 1063-6919

  90. Detecting Building-level Changes of a City Using Street Images and a 2D City Map Peer-reviewed

    Daiki Tetsuka, Takayuki Okatani

    2015 IEEE Winter Conference on Applications of Computer Vision (WACV) 349-356 2015

    DOI: 10.1109/WACV.2015.53  

  91. Separation of Reflection Components by Sparse Non-negative Matrix Factorization Peer-reviewed

    Yasuhiro Akashi, Takayuki Okatani

    COMPUTER VISION - ACCV 2014, PT V 9007 611-625 2015

    DOI: 10.1007/978-3-319-16814-2_40  

    ISSN: 0302-9743

  92. Understanding Convolutional Neural Networks in Terms of Category-Level Attributes Peer-reviewed

    Makoto Ozeki, Takayuki Okatani

    COMPUTER VISION - ACCV 2014, PT II 9004 362-375 2015

    DOI: 10.1007/978-3-319-16808-1_25  

    ISSN: 0302-9743

  93. Massive City-Scale Surface Condition Analysis Using Ground and Aerial Imagery Peer-reviewed

    Ken Sakurada, Takayuki Okatani, Kris M. Kitani

    COMPUTER VISION - ACCV 2014, PT I 9003 49-64 2015

    DOI: 10.1007/978-3-319-16865-4_4  

    ISSN: 0302-9743

  94. Creating multi-viewpoint panoramas of streets with sparsely located buildings Peer-reviewed

    Takayuki Okatani, Jun Yanagisawa, Daiki Tetsuka, Ken Sakurada, Koichiro Deguchi

    Springer Tracts in Advanced Robotics 92 65-79 2014

    DOI: 10.1007/978-3-642-40686-7_5  

    ISSN: 1610-7438 1610-742X

  95. Improving accuracy of planar tracking by resolving resolution inconsistencies Peer-reviewed

    Tomohiro Ushiki, Eisuke Ito, Takayuki Okatani

    IPSJ Transactions on Computer Vision and Applications 5 109-113 2013/07

    DOI: 10.2197/ipsjtcva.5.109  

    ISSN: 1882-6695

  96. Detecting Changes in 3D Structure of a Scene from Multi-view Images Captured by a Vehicle-mounted Camera Peer-reviewed

    Ken Sakurada, Takayuki Okatani, Koichiro Deguchi

    2013 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) 137-144 2013

    DOI: 10.1109/CVPR.2013.25  

    ISSN: 1063-6919

  97. Discrete MRF Inference of Marginal Densities for Non-uniformly Discretized Variable Space Peer-reviewed

    Masaki Saito, Takayuki Okatani, Koichiro Deguchi

    2013 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) 57-64 2013

    DOI: 10.1109/CVPR.2013.15  

    ISSN: 1063-6919

  98. Application of the Mean Field Methods to MRF Optimization in Computer Vision Peer-reviewed

    Masaki Saito, Takayuki Okatani, Koichiro Deguchi

    2012 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) 1680-1687 2012

    DOI: 10.1109/CVPR.2012.6247862  

    ISSN: 1063-6919

  99. Optimal Integration of Photometric and Geometric Surface Measurements Using Inaccurate Reflectance/Illumination Knowledge Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    2012 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) 254-261 2012

    DOI: 10.1109/CVPR.2012.6247683  

    ISSN: 1063-6919

  100. Recognizing Surface Qualities from Natural Images based on Learning to Rank Peer-reviewed

    Takashi Abe, Takayuki Okatani, Koichiro Deguchi

    2012 21ST INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR 2012) 3712-3715 2012

    ISSN: 1051-4651

  101. Optimum method for real-time reconstruction of sensor surface in total-internal-reflection based tactile sensor. Peer-reviewed

    Ryosuke Taira, Satoshi Saga, Takayuki Okatani, Koichiro Deguchi

    IEEE World Haptics Conference, WHC 2011, 21-24 June 2011, Istanbul, Turkey 329-334 2011

    Publisher: IEEE Computer Society

    DOI: 10.1109/WHC.2011.5945507  

  102. Flexible Online Calibration for a Mobile Projector-Camera System Peer-reviewed

    Daisuke Abe, Takayuki Okatani, Koichiro Deguchi

    COMPUTER VISION - ACCV 2010, PT IV 6495 565-579 2011

    DOI: 10.1007/978-3-642-19282-1_45  

    ISSN: 0302-9743

  103. Hand Gesture Recognition using Histogram of Oriented Gradients and Partial Least Squares Regression Peer-reviewed

    Arindam Misra, Abe Takashi, Takayuki Okatani, Koichiro Deguchi

    Proceedings of IAPR Conference on Machine Vision Applications 479-482 2011

  104. Sensing method of total-internal-reflection-based tactile sensor Peer-reviewed

    Momotaro Koike, Satoshi Saga, Takayuki Okatani, Koichiro Deguchi

    2011 IEEE World Haptics Conference, WHC 2011 615-619 2011

    DOI: 10.1109/WHC.2011.5945556  

  105. Accurate and Robust Planar Tracking based on a Model of Image Sampling and Reconstruction Process Peer-reviewed

    Eisuke Ito, Takayuki Okatani, Koichiro Deguchi

    2011 10TH IEEE INTERNATIONAL SYMPOSIUM ON MIXED AND AUGMENTED REALITY (ISMAR) 2011

    DOI: 10.1109/ISMAR.2011.6092364  

    ISSN: 1554-7868

  106. Efficient Algorithm for Low-rank Matrix Factorization with Missing Components and Performance Comparison of Latest Algorithms Peer-reviewed

    Takayuki Okatani, Takahiro Yoshida, Koichiro Deguchi

    2011 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV) 842-849 2011

    DOI: 10.1109/ICCV.2011.6126324  

    ISSN: 1550-5499

  107. 3D reconstruction of reflective surface on reflection type tactile sensor using constraints of geometrical optics

    Ryosuke Taira, Satoshi Saga, Takayuki Okatani, Koichiro Deguchi

    Proceedings of the SICE Annual Conference 3144-3149 2010

    Publisher: Society of Instrument and Control Engineers (SICE)

  108. A projector-camera system for high-quality synthesis of virtual reflectance on real object surfaces Peer-reviewed

    Tomoya Okazaki, Takayuki Okatani, Koichiro Deguchi

    IPSJ Transactions on Computer Vision and Applications 2 71-83 2010

    DOI: 10.2197/ipsjtcva.2.71  

    ISSN: 1882-6695

  109. Easy calibration of a multi-projector display system Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    International Journal of Computer Vision 85 (1) 1-18 2009/04

    DOI: 10.1007/s11263-009-0242-0  

  110. 被写界深度ぼけの提示により奥行き感を強化する注視反応型ディスプレイ Peer-reviewed

    岡谷貴之, 石澤昴, 出口光一郎

    電子情報通信学会論文誌 J92-D (8) 1298-1307 2009/03

  111. Study of Image Quality of Superimposed Projection Using Multiple Projectors Peer-reviewed

    Takayuki Okatani, Mikio Wada, Koichiro Deguchi

    IEEE TRANSACTIONS ON IMAGE PROCESSING 18 (2) 424-429 2009/02

    DOI: 10.1109/TIP.2008.2008899  

    ISSN: 1057-7149

  112. Modeling effects of urban spatial structure on urban thermal environment - Integrating satellite image processing and metric model analysis

    Nguyen Thi Mai Dung, Koichiro Deguchi, Takayuki Okatani

    30th Asian Conference on Remote Sensing 2009, ACRS 2009 1 590-596 2009

  113. Shape Reconstruction by Combination of Structured-Light Projection and Photometric Stereo Using a Projector-Camera System High Quality Reproduction of a Virtual Reflectance Property on a Real Object Surface Peer-reviewed

    Tomoya Okazaki, Takayuki Okatani, Koichiro Deguchi

    ADVANCES IN IMAGE AND VIDEO TECHNOLOGY, PROCEEDINGS 5414 410-422 2009

    ISSN: 0302-9743

  114. On Bias Correction for Geometric Parameter Estimation in Computer Vision Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    CVPR: 2009 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, VOLS 1-4 959-966 2009

    DOI: 10.1109/CVPRW.2009.5206722  

    ISSN: 1063-6919

  115. Improving Accuracy of Geometric Parameter Estimation Using Projected Score Method Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    2009 IEEE 12TH INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV) 1733-1740 2009

    DOI: 10.1109/ICCV.2009.5459388  

    ISSN: 1550-5499

  116. Absolute phase measurements using geometric constraints between multiple cameras and projectors Peer-reviewed

    Rui Ishiyama, Shizuo Sakamoto, Johji Tajima, Takayuki Okatani, Koichiro Deguchi

    Applied Optics 46 (17) 3528-3539 2007/06/10

    Publisher: OSA - The Optical Society

    DOI: 10.1364/AO.46.003528  

    ISSN: 1539-4522 1559-128X

    eISSN: 1539-4522

  117. On the Wiberg algorithm for matrix factorization in the presence of missing components Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    INTERNATIONAL JOURNAL OF COMPUTER VISION 72 (3) 329-337 2007/05

    DOI: 10.1007/s11263-006-9785-5  

    ISSN: 0920-5691

  118. 3D spatial information extraction from ALOS PRISM triple image

    Nguyen Thi Mai Dung, Koichiro Deguchi, Takayuki Okatani

    28th Asian Conference on Remote Sensing 2007, ACRS 2007 2 1199-1203 2007

  119. A Physics-Based Imaging Model of Scanning Electron Microscopes Peer-reviewed

    Kousuke Kamada, Takayuki Okatani, Koichiro Deguchi

    Proceedings of IAPR Conference on Machine Vision Applications 102-105 2007

  120. Variational Bayes based approach to robust subspace learning Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    2007 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, VOLS 1-8 1004-+ 2007

    DOI: 10.1109/CVPR.2007.383101  

    ISSN: 1063-6919

  121. Estimating scale of a scene from a single image based on defocus blur and scene geometry Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    2007 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, VOLS 1-8 1890-+ 2007

    DOI: 10.1109/CVPR.2007.383217  

    ISSN: 1063-6919

  122. Precise 3-D measurement using uncalibrated pattern projection Peer-reviewed

    Rui Ishiyama, Takayuki Okatani, Koichiro Deguchi

    2007 IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING, VOLS 1-7 (1) 225-228 2007

    DOI: 10.1109/ICIP.2007.4378932  

    ISSN: 1522-4880

  123. HHMM based recognition of human activity Peer-reviewed

    Daiki Kawanaka, Takayuki Okatani, Koichiro Deguchi

    IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS E89D (7) 2180-2185 2006/07

    DOI: 10.1093/ietisy/e89-d.7.2180  

    ISSN: 1745-1361

  124. Video synchronization based on co-occurrence of appearance changes in video sequences Peer-reviewed

    Manabu Ushizaki, Takayuki Okatani, Kochiro Deguchi

    18TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION, VOL 3, PROCEEDINGS 71-+ 2006

    DOI: 10.1109/ICPR.2006.1174  

    ISSN: 1051-4651

  125. The importance of gaze control mechanism on vision-based motion control of a biped robot Peer-reviewed

    Shun Ushida, Kousuke Yoshimi, Takayuki Okatani, Koichiro Deguchi

    2006 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS, VOLS 1-12 4447-+ 2006

    DOI: 10.1109/IROS.2006.282079  

  126. Autocalibration of an ad hoc construction of multi-projector displays Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition 2006 2006

    DOI: 10.1109/CVPRW.2006.35  

    ISSN: 1063-6919

  127. Autocalibration of a projector-camera system Peer-reviewed

    T Okatani, K Deguchi

    IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 27 (12) 1845-1855 2005/12

    DOI: 10.1109/TPAMI.2005.235  

    ISSN: 0162-8828

    eISSN: 1939-3539

  128. 行動と画像変化を結びつけたロボットの空間認識 Peer-reviewed

    中川智洋, 岡谷貴之, 出口光一郎

    日本ロボット学会誌 23 (3) 90-97 2005/03

    Publisher: The Robotics Society of Japan

    DOI: 10.7210/jrsj.23.362  

    ISSN: 0289-1824

    More details Close

    In this paper, we propose an active vision strategy for the construction of a 3D map in a robot brain from its stereo eye images. We show that the combination of the robot action and image change caused by the action improves the accuracy of the map. If the robot stereo cameras have been accurately calibrated, the obtained reconstruction of the static scene stays stationary. If the calibration is not accurate, the reconstructed scene changes according to the robot action. Therefore, we utilize this change to correct the parameters of the stereo cameras and the 3D map so as to obtain stationary scene under the action. We show the feasibility of this idea as an auto-calibration of robot vision with some simulation experiments and implementation on actual robots.

  129. HHMM based recognition of human activity from motion trajectories in image sequences

    Daiki Kawanaka, Shun Ushida, Takayuki Okatani, Koichiro Deguchi

    Proceedings of the 9th IAPR Conference on Machine Vision Applications, MVA 2005 578-581 2005

  130. Markerless augmented reality using planar structures

    Hajime Tomizawa, Takayuki Okatani, Koichiro Deguchi

    Proceedings of the SICE Annual Conference 1066-1070 2005

  131. A method for 3D shape reconstruction from SEM images

    Kosuke Kamada, Koichiro Deguchi, Takayuki Okatani

    Proceedings of the SICE Annual Conference 650-653 2005

  132. Reconstruction of the spatial motion and posture of tools from stereo image sequence

    Manabu Ushizaki, Takayuki Okatani, Koichiro Deguchi

    Proceedings of the SICE Annual Conference 644-648 2005

  133. Motion Stabilization of Biped Robot by Gaze Control Peer-reviewed

    Shota Takizawa, Shun Ushida, Takayuki Okatani, Koichiro Deguchi

    Proceedings of IAPR Conference on Machine Vision Applications 324-327 2005

  134. 2DOF motion stabilization of biped robot by gaze control strategy Peer-reviewed

    S Takizawa, S Ushida, T Okatani, K Deguchi

    2005 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS, VOLS 1-4 3809-3814 2005

    DOI: 10.1109/IROS.2005.1545562  

  135. Contrast enhancement of fundus images using regional histograms for medical diagnosis

    Tatsuya Shimahara, Takayuki Okatani, Koichiro Deguchi

    Proceedings of the SICE Annual Conference 411-414 2004

  136. High-speed and high-precision optical flow detection for real-time motion segmentation Peer-reviewed

    H Ishiyama, T Okatani, K Deguchi

    SICE 2004 ANNUAL CONFERENCE, VOLS 1-3 1202-1205 2004

  137. A probabilistic approach to linear subspace fitting for computer vision problems Peer-reviewed

    Takayuki Okatani

    IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops 2004- (January) 2004

    Publisher: IEEE Computer Society

    DOI: 10.1109/CVPR.2004.286  

    ISSN: 2160-7516 2160-7508

  138. Active construction of 3D map in robot by combining motion and perceived images Peer-reviewed

    Tomohiro Nakagawa, Takayuki Okatani, Koichiro Deguchi

    Proceedings of Asian Conference on Computer Vision 1 563-568 2004

  139. Object tracking by the mean-shift of regional color distribution combined with the particle-filter algorithm Peer-reviewed

    K Deguchi, O Kawanaka, T Okatani

    PROCEEDINGS OF THE 17TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION, VOL 3 506-509 2004

    DOI: 10.1109/ICPR.2004.1334577  

    ISSN: 1051-4651

  140. A Color-based Probabilistic Tracking by Using Graphical Models Peer-reviewed

    Y. Sato, T. Okatani, K. Deguchi

    Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems 631-636 2004

  141. A color-based tracking by Kalman particle filter Peer-reviewed

    Y Satoh, T Okatani, K Deguchi

    PROCEEDINGS OF THE 17TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION, VOL 3 502-505 2004

    DOI: 10.1109/ICPR.2004.1334576  

    ISSN: 1051-4651

  142. 実環境での物体認識のための距離画像マッチング Peer-reviewed

    長尾影洋, 岡谷貴之, 出口光一郎

    計測自動制御学会論文誌 40 (1) 1-8 2004/01

    Publisher: The Society of Instrument and Control Engineers

    DOI: 10.9746/sicetr1965.40.10  

    ISSN: 0453-4654

    More details Close

    Object recognition is very important task in the field of computer vision. In this paper, we present a range image matching for the object recognition in real scene. The matching is the process of comparing a range image with a stored shape data. We first discuss that the difference of resolutions of them must be considered for the range image matching. Then, we analyze the apparent object shape change of the range image due to the measurement distance in the relation to the property of the range finder. We claim that the apparent shape change can be modelled by a smoothing process of the object surface. We propose a method of matching after shape smoothing of the stored data to cope with has a very low resolution of the range image. We also propose a generation of a surface model of the given data, considering the resolution difference of a newly obtained data of the target object. Finally, we show several experimental results which promises the feasibility of the proposal method.

  143. Comparison of the feature point tracking method in image sequences Peer-reviewed

    Y Naito, T Okatani, K Deguchi

    SICE 2003 ANNUAL CONFERENCE, VOLS 1-3 1326-1331 2003

  144. An extraction method of moving objects by shape energy functions with prior knowledge Peer-reviewed

    Y Satoh, T Okatani, K Deguchi

    SICE 2003 ANNUAL CONFERENCE, VOLS 1-3 1320-1325 2003

  145. Tracking multiple three-dimensional motions by using modified condensation algorithm and multiple images Peer-reviewed

    K Hamasaki, T Nakajima, T Okatani, K Deguchi

    IROS 2003: PROCEEDINGS OF THE 2003 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS, VOLS 1-4 236-241 2003

  146. Autocalibration of a projector-sereen-camera system: Theory and algorithm for screen-to-camera homography estimation Peer-reviewed

    T Okatani, K Deguchi

    NINTH IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION, VOLS I AND II, PROCEEDINGS 774-781 2003

  147. Binocular motion tracking by gaze fixation control and three-dimensional shape reconstruction Peer-reviewed

    Y Satoh, T Okatani, K Deguchi

    ADVANCED ROBOTICS 17 (10) 1057-1072 2003

    DOI: 10.1163/156855303322554427  

    ISSN: 0169-1864

    eISSN: 1568-5535

  148. Toward a statistically optimal method for estimating geometric relations from noisy data: Cases of linear relations Peer-reviewed

    T Okatani, K Deguchi

    2003 IEEE COMPUTER SOCIETY CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, VOL 1, PROCEEDINGS 432-439 2003

    ISSN: 1063-6919

  149. 3D object surface generation from range point data for object shape model matching Peer-reviewed

    K Nagao, T Okatani, K Deguchi

    SICE 2002: PROCEEDINGS OF THE 41ST SICE ANNUAL CONFERENCE, VOLS 1-5 1416-1419 2002

  150. Is There Room for Improving Estimation Accuracy of the Structure and Motion Problem? Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    Workshop of Statistical Method in Video Processing 2002

  151. Robust estimation of camera translation between two images using a camera with a 3D orientation sensor Invited Peer-reviewed

    T Okatani, K Deguchi

    16TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION, VOL I, PROCEEDINGS 275-278 2002

    ISSN: 1051-4651

  152. 3-Dimensional Tracking of Multiple Object Motions from Multi-view Image by Mixed-State CONDENSATION Algorithm Peer-reviewed

    Koichiro Deguchi, Koji Hamasaki, Taira Nakajima, Takayuki Okatani

    Proceedings of IAPR Workshop on Machine Vision Applications 408-411 2002

  153. Theory of Nuisance Parameter Problems in Computer Vision: Several Examples of Application to Linear Cases Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    Proceedings of the Workshop on Science of Vision 94-100 2002

  154. A motion tracking by extracting 3D feature of moving objects with binocular cooperative fixation Peer-reviewed

    Y Satoh, T Nakagawa, T Okatani, K Deguchi

    2002 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS, VOLS 1-3, PROCEEDINGS 13-18 2002

  155. カメラモデルの段階的切り換えによる多視点カメラシステムのセルフキャリブレーション Peer-reviewed

    岡谷貴之, 久保田俊作, 出口光一郎

    情報処理学会論文誌:コンピュータビジョンとイメージメディア 43 (11) 33-42 2002

  156. On uniqueness of solutions of the three-light-source photometric stereo: Conditions on illumination configuration and surface reflectance Peer-reviewed

    T Okatani, K Deguchi

    COMPUTER VISION AND IMAGE UNDERSTANDING 81 (2) 211-226 2001/02

    DOI: 10.1006/cviu.2000.0887  

    ISSN: 1077-3142

  157. On photometric aspects of catadioptric cameras Peer-reviewed

    T Okatani, K Deguchi

    2001 IEEE COMPUTER SOCIETY CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, VOL 1, PROCEEDINGS 1106-1113 2001

    ISSN: 1063-6919

  158. 反射鏡を利用した全方位カメラの画像照度について Peer-reviewed

    岡谷貴之, 出口光一郎

    情報処理学会論文誌:コンピュータビジョンとイメージメディア 42 (13) 71-80 2001

  159. Closed form solution of local shape from shading at critical points Peer-reviewed

    T Okatani, K Deguchi

    INTERNATIONAL JOURNAL OF COMPUTER VISION 40 (2) 169-178 2000/11

    DOI: 10.1023/A:1026510403689  

    ISSN: 0920-5691

    eISSN: 1573-1405

  160. 非ランバート性拡散反射に対する等高線の発展による陰影からの形状復元 Peer-reviewed

    OKATANI Takayuki

    情報処理学会論文誌 41 (1) 92-100 2000

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 1882-7764

    More details Close

    The problem of recovering shape from shading is formulated as solving the first order partial differential equation. Besides the classical method of the characteristic equation, a group of methods that computes the solution by tracking its equal hight contour has been proposed. But all of them assumed the Lambertian reflectance. This paper extends the methods to general diffuse reflectance that is not necessarily Lambertian. We show a condition that the reflectance map should satisfy so that the methods can be used.

  161. Estimation of illumination distribution using a specular sphere

    T Okatani, K Deguchi

    15TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION, VOL 3, PROCEEDINGS 592-595 2000

    ISSN: 1051-4651

  162. Recovering Camera Motion from Image Sequence Based on Registration of Silhouette Cones : Shape from Silhouette Using a Mobile Camera with a Gyro Sensor Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    IAPR Workshop on Machine Vision Applications 451-454 2000

  163. Estimation of illumination distribution using a specular sphere Peer-reviewed

    T Okatani, K Deguchi

    15TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION, VOL 3, PROCEEDINGS 592-595 2000

    ISSN: 1051-4651

  164. Yet Another Appearance-Based Method for Pose Estimation Based on a Linear Model Peer-reviewed

    Takayuki OKATANI, Koichiro DEGUCHI

    Proceedings of IAPR Conference on Machine Vision Applications 2000

  165. Computation of the sign of the Gaussian curvature of a surface from multiple unknown illumination images without knowledge of the reflectance property Peer-reviewed

    T Okatani, K Deguchi

    COMPUTER VISION AND IMAGE UNDERSTANDING 76 (2) 125-134 1999/11

    DOI: 10.1006/cviu.1999.0792  

    ISSN: 1077-3142

  166. 陰影画像における特異点に関する等濃度線の不変性について Peer-reviewed

    OKATANI Takayuki

    情報処理学会論文誌 40 (9) 3446-3454 1999

  167. 陰影と反射率分布図の整合性について

    岡谷貴之, 出口光一郎

    情報処理学会研究報告 117 (6) 39-46 1999

  168. 複数の画像間の陰影差に基づく曲画のガウス曲率の符号の決定 Peer-reviewed

    OKATANI Takayuki

    情報処理学会論文誌 39 (6) 1965-1972 1998

  169. On identification of singular points using photometric invariants for global shape from shading problem Peer-reviewed

    T Okatani, K Deguchi

    FOURTEENTH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION, VOLS 1 AND 2 1787-1790 1998

    ISSN: 1051-4651

  170. Shape reconstruction from an endoscope image by shape from shading technique for a point light source at the projection center Peer-reviewed

    T Okatani, K Deguchi

    COMPUTER VISION AND IMAGE UNDERSTANDING 66 (2) 119-131 1997/05

    DOI: 10.1006/cviu.1997.0613  

    ISSN: 1077-3142

  171. 内視鏡画像からの陰影に基づく立体形状復元 Peer-reviewed

    OKATANI Takayuki

    計測自動制御学会論文集 33 (10) 1035-1042 1997

    Publisher: The Society of Instrument and Control Engineers

    DOI: 10.9746/sicetr1965.33.1035  

    ISSN: 0453-4654

    More details Close

    A new method for reconstructing three dimensional object from an endoscope image is presented. The proposed method uses image shading generated by a light source at the endoscope head. The problem of reconstructing three dimensional shape of an object from its shading has been widely studied in the literature of computer vision research. There have been many significant developments made in this area. However, most of the conventional methods are not applicable to this case, since they are based on the assumption that the light source is far from the object surface. The light source of an endoscope is located at its head, so that it is near the object surface. In this case, the image brightness depends on not only the surface gradients but also the distance from the light source to each point on the surface. To deal with this difficulty, we employ an optical system having a point light source at the projection center, which well approximates the imaging system of the endoscope. In addition, we introduce an equal-distance contour, which is a curve composed of surface points at an equal distance from the light source. Propagating these contours using the image shading, we reconstruct the object surface. We use the level-set method for numerical computation. Experimental results for real medical endoscope images show feasibility of this method.

  172. Determination of sign of gaussian curvature of surface having general reflectance property Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 1351 56-63 1997

    Publisher: Springer Verlag

    DOI: 10.1007/3-540-63930-6_104  

    ISSN: 1611-3349 0302-9743

  173. On the classification of singular points for the global shape from shading problem: A study of the constraints imposed by isophotes Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 1351 48-55 1997

    Publisher: Springer Verlag

    DOI: 10.1007/3-540-63930-6_103  

    ISSN: 1611-3349 0302-9743

  174. Shape reconstruction from an endoscope image by shape-from-shading technique for a point light source at the projection center Peer-reviewed

    K Deguchi, T Okatani

    PROCEEDINGS OF THE IEEE WORKSHOP ON MATHEMATICAL METHODS IN BIOMEDICAL IMAGE ANALYSIS 290-298 1996

  175. Reconstructing shape from shading with a point light source at the projection center: Shape reconstruction from an endoscope image Peer-reviewed

    Takayuki Okatani, Koichiro Deguchi

    Proceedings - International Conference on Pattern Recognition 1 830-834 1996

    Publisher: Institute of Electrical and Electronics Engineers Inc.

    DOI: 10.1109/ICPR.1996.546140  

    ISSN: 1051-4651

Show all ︎Show first 5

Misc. 74

  1. ImPACT TRC Thin Serpentine Robot Platform: Air-floating-type Active Scope Camera Integrated with Multiple Sensory Functions

    昆陽雅司, 安部祐一, 永野光, 坂東宜昭, 山崎公俊, 岡谷貴之, 糸山克寿, 奥乃博, 田所諭

    システム制御情報学会研究発表講演会講演論文集(CD-ROM) 62nd ROMBUNNO.141‐5 2018/05/16

    Publisher: システム制御情報学会

  2. 感覚機能統合型能動スコープカメラの改良と瓦礫フィールドへの適用

    山内悠, 安部祐一, 永野光, 昆陽雅司, 坂東宜昭, 山崎公俊, 糸山克寿, 猿渡洋, 岡谷貴之, 奥乃博, 田所諭

    計測自動制御学会システムインテグレーション部門講演会(CD-ROM) 18th ROMBUNNO.1C6‐09 2017/12/20

  3. Development of active scope camera with sensory functions

    Ambe Yuichi, Tadokoro Satoshi, Bando Yoshiaki, Nagano Hikaru, Konyo Masashi, Yamazaki Kimitoshi, Itoyama Katsutoshi, Saruwatari Hiroshi, Okatani Takayuki, G. Okuno Hiroshi

    The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec) 2017 (0) 1P2-P01 2017/05/09

    Publisher: 一般社団法人 日本機械学会

    DOI: 10.1299/jsmermd.2017.1p2-p01  

    eISSN: 2424-3124

    More details Close

    <p>We had developed an active scope camera: the robot video scope that can move by itself to probe narrow gaps for rescue missions. However, to investigate the interiors of collapsed houses effectively, not only the high mobility but also the sensors, such as vision, audition and haptics should be installed on the robot. According to this, we have developed an active scope camera with the sensory functions and demonstrated these functions in the test field which imitates collapsed houses. In the demonstration, we mainly showed four functions. (1)Detection of collision and it's presentation to the operator, (2)Searching operation assist with the processed vision data, (3)Shape estimation of the robot with sound and IMU data, and (4)Voice enhancement to detect victim's voice.</p>

  4. 深層学習とその画像応用に関する最新動向

    岡谷貴之

    日本画像学会誌 56 (2) 157-162 2017/02

    Publisher: 日本画像学会

    DOI: 10.11370/isj.56.157  

    ISSN: 1344-4425

  5. 索状ロボットのためのロバストな視覚SLAMの実現

    清水寛太, XU Liang, 岡谷貴之, 安部祐一, 昆陽雅司, 田所諭

    日本機械学会ロボティクス・メカトロニクス講演会講演論文集(CD-ROM) 2017 2017

    ISSN: 2424-3124

  6. 救助犬に搭載した全方位カメラを用いた視覚オドメトリ

    矢田亮太郎, 清水寛太, 岡谷貴之, 大野和則, 山口竣平, 濱田龍之介, 田所諭

    日本機械学会ロボティクス・メカトロニクス講演会講演論文集(CD-ROM) 2017 2017

    ISSN: 2424-3124

  7. ImPACT TRC太索状ロボット研究開発の現状

    松野文俊, 伊藤一之, 亀川哲志, 田中基康, 有泉亮, 奥乃博, 大道武生, 鈴木陽介, 岡谷貴之, 多田隈建二郎

    日本ロボット学会学術講演会予稿集(CD-ROM) 34th ROMBUNNO.3C2‐05 2016/09/07

  8. ディープラーニング

    岡谷貴之

    映像メディア学会誌 68 (6) 466-471 2016/07/30

    Publisher: The Institute of Image Information and Television Engineers

    DOI: 10.3169/itej.68.466  

    ISSN: 1342-6907

  9. Research Trend in Deep Learning for Visual Recognition : Advances of Convolutional Neural Networks and Their Use(<Special Issue>Research Frontiers in Neural Network)

    Takayuki Okatani, Graduate School of Information Sciences Tohoku University

    31 (2) 169-179 2016/03/01

    Publisher: 人工知能学会 ; 2014-

    ISSN: 2188-2266

  10. Revolutions in Image Recognition:2. Visual Recognition by Deep Learning : Ability and Limitation of Convolutional Neural Networks

    56 (7) 634-639 2015/06/15

    Publisher: 情報処理学会 ; 1960-

    ISSN: 0447-8053

  11. 画像認識の現状認識:ディープラーニングの成功と限界?

    岡谷 貴之

    情報処理学会研究報告. SLP, 音声言語情報処理 2015 (11) 1-1 2015/05/18

    Publisher: 一般社団法人情報処理学会

    More details Close

    一枚の画像からそこに写る物体の名前を答える物体カテゴリ認識は,コンピュータビジョンの長年の課題であったが,ディープラーニングの登場,正確には畳込みニューラルネットの 「再発見」 により大きく進展し,人の視覚と比肩するほどまでになった.一方,その他の多くの画像認識のタスク,例えば特定の物体を画像中で位置決めする 「物体検出」 や,人のポーズ認識,さらには動画像を使った人の行動認識に対しては,物体カテゴリ認識ほどはディープラーニングは成功していない.従来法を超えるという意味では一定の有効性が確認されているものの,これらのタスクでの性能は,人の視覚まではまだ相当の隔たりがある.また,物体カテゴリ認識あるいはその類似タスクに極めて高い精度を発揮する畳込みニューラルネットだが,なぜそんなに高い性能が出るのかの理解が追い付いていない.本講演では,この辺りの現状についてまとめ,今後を占うことを試みたい.

  12. Four-dimensional City Modeling using Vehicular Imagery

    IPSJ SIG Notes. CVIM 2015 (35) 1-16 2015/05/11

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    We propose a novel method for four-dimensional city modeling using vehicular imagery. Motivation behind this study is to estimate and visualize damage and recovery process of tsunami-damaged area. To conduct the research, we have been recording the images in tsunami-damaged area periodically driving a car on which omnidirectional camera is mounted. To estimate temporal changes of a wide area using vehicular imagery, there are many challenges to overcome, for example, limitation of camera viewpoint, coverage of vehicular imagery and illumination condition. Furthermore, a 3D model is not always available for every city, and the scene images do often not have sufficient visual features to perform precise registration. In the case of a wide-area disaster, it is computationally prohibitive to reconstruct the three-dimensional structure of entire areas. To overcome these difficulties, we propose 2D, 3D and object-based change detection methods. The 2D method detects scene change from an image pair using visual features of convolutional neural network. The 3D method estimates structural changes of scene from images taken at multiple viewpoints even if there is depth ambiguity of a scene. The object-based method estimates land surface condition of an entire city integrating aerial and street-view imagery, which are taken at vastly different viewpoints. The experimental results show that our methods can accurately and effectively estimate temporal changes of a city.

  13. 深層学習について

    岡谷貴之

    日本ロボット学会誌 33 (2) 92-96 2015/04/15

    DOI: 10.7210/jrsj.33.92  

  14. ディープラーニングと画像認識 : 基礎と最近の動向 (特集 ニューロサイエンスと数理モデリング)

    岡谷 貴之

    オペレーションズ・リサーチ 60 (4) 198-204 2015/04

    Publisher: 公益社団法人日本オペレーションズ・リサーチ学会

    ISSN: 0030-3674

    More details Close

    ディープラーニングは近年,人工知能の色々な分野で大きな成功を収めつつあり,その高い性能は広く知られるようになった.本稿ではディープラーニングの画像への応用,中でも画像認識に焦点を絞り,そこで欠かせない存在となっている畳込みニューラルネットについて,その技術的な基礎から最近の研究の動向までを概観する.

  15. 畳込みニューラルネットワークを用いたシーン変化の検出

    櫻田 健, 岡谷 貴之

    研究報告コンピュータビジョンとイメージメディア(CVIM) 2015 (2) 1-8 2015/02/27

    Publisher: 一般社団法人情報処理学会

    ISSN: 0919-6072

    More details Close

    本稿はグリッド特徴を利用して各時刻一枚ずつの画像ペアからシーンの変化を検出する手法を提案する.自動のシーン変化の検出は都市管理や,災害の復旧,復興,減災において有用である.本研究の背景として,車載カメラの画像を用いて津波被災地の街並みの変化を可視化する目的がある.既存のシーン変化の検出手法は,街並みの 3 次元モデルや,異なる時刻のデータ間におけるピクセルレベルの位置合わせを必要とした.そのため,正確な 3 次元モデルが得られない本研究のようなケースには適用することが困難である.さらに,広域の 3 次元モデルの復元や,ピクセルレベルでの位置合わせは膨大な計算量を必要とする.これらの問題を解決するために,本研究では畳込みニューラルネットワークとスーパーピクセルセグメンテーションを統合した新しいシーン変化の検出方法を提案する.提案手法はシーンの 3 次元モデルやピクセルレベルでの位置合わせを必要とせず,計算量を大幅に削減できるため,津波被災地全体のシーン変化の推定を可能とする.

  16. Visual Recognition of Surface Qualities of Objects : Current Research and Future Prospects

    OKATANI Takayuki

    Technical report of IEICE. Multimedia and virtual environment 114 (410) 205-205 2015/01/15

    Publisher: The Institute of Electronics, Information and Communication Engineers

  17. Deep Learning for Image Recognition(<Survey Papers>Deep Learning (4))

    Takayuki Okatani, Graduate School of Information Sciences Tohoku University

    28 (6) 962-974 2013/11/01

    Publisher: 人工知能学会

    ISSN: 0912-8085

  18. Learning Features by Deep Networks and Its Application to Image Recognition

    Okatani Takayuki

    IEICE technical report. 113 (146) 69-69 2013/07/11

    Publisher: The Institute of Electronics, Information and Communication Engineers

  19. On the Volumetric Reconstruction of Linear Objects from Multiple Visual Scenes

    YOSHIDA TAKAHIRO, SAGA SATOSHI, OKATANI TAKAYUKI, DEGUCHI KOICHIRO

    2013 (30) 1-6 2013/05/23

    Publisher: 情報処理学会

    ISSN: 0919-6072

  20. ディープラーニング (パターン認識・メディア理解)

    岡谷 貴之, 齋藤 真樹

    電子情報通信学会技術研究報告 : 信学技報 112 (385) 111-127 2013/01/23

    Publisher: 一般社団法人電子情報通信学会

    ISSN: 0913-5685

    More details Close

    ディープラーニングは,多層のニューラルネットワーク(以下NN)を用いる機械学習の方法論である.最近,これに基づく方法群が,画像認識のベンチマークテストで従来の記録を次々に塗り替えた他,音声認識やその他の学習・認識の問題に応用されて同様に高い性能を示すに至り,幅広い関心を集めつつある.本稿では,このディープラーニングについて,画像認識への応用を念頭に,現時点で知られている様々な方法をなるべく網羅的に説明する.具体的には, NNの基本構成から, Convolutional NN, プーリング,局所コントラスト正規化,教師なし学習であるプレトレーニング,オートエンコーダ,スパースな特徴表現を可能にする複数の方法, Restrlcted Boltzmann MachineやDeep Belief Networkなどの生成モデルに基づくディープラーニングの各手法を,それぞれ説明する.またディープラーニングのためのソフトウェアライブラリにも触れる.

  21. 車載カメラで撮影した画像列と市街地地図情報を用いた建築物の出現・消失の推定 (パターン認識・メディア理解)

    手塚 大基, 岡谷 貴之, 出口 光一郎

    電子情報通信学会技術研究報告 : 信学技報 112 (385) 51-57 2013/01/23

    Publisher: 一般社団法人電子情報通信学会

    ISSN: 0913-5685

    More details Close

    本研究では車載カメラで走行しながら市街地を撮影した画像群及び,建築物の輪郭情報が記載された住宅地図を使って,両者の間の不整合を自動的に特定する方法を述べる.例えば,住宅地図上には存在するが,実際には消失した建築物の自動判定か行なえる.具体的には画像群からシーンの3次元復元を行い,得られた3次元情報と地図上の建築物との比較を行い,建築物の存在を確率的に推定する.東日本大震災で被災した市街地を撮影して得た被画像と,被災前の住宅地図を使って実験を行ったので,その結果を報告する.

  22. ロバストバンドル調整を用いた車載ステレオカメラのセルフキャリブレーション (パターン認識・メディア理解)

    深澤 至貴, 岡谷 貴之, 出口 光一郎, 木村 宣隆

    電子情報通信学会技術研究報告 : 信学技報 112 (385) 59-66 2013/01/23

    Publisher: 一般社団法人電子情報通信学会

    ISSN: 0913-5685

    More details Close

    車載ステレオカメラのセルフキャリブレーション,つまり車両が一般道を通常走行する際に得られる画像のみを用いたキャリブレーションの方法を述べる.具体的には,各カメラの内部パラメータおよびレンズ歪がオフラインのキャリブレーションて既知であるとし,ステレオカメラを構成する左右カメラの間の相対姿勢を推定する.この場合,自然のシーンを撮影した左右カメラの画像間でいかに外れ値のない点対応を得られるか,あるいは混入した外れ値を除去できるかが鍵となる.走行中の連続画像間およひ左右カメラの画像間て特徴点を追跡・対応付けてその軌跡を得,それらをすべて使用して完全な3次元復元(SfM)を行い,ロバストバントル調整によって全体最適化をすることで,得た点対応に混入する外れ値を効果的に取り除き,高精度に目的パラメータを推定する.実験結果は,この方法がステレオカメラのキャリブレーションに有効であることを示した.

  23. ロバストバンドル調整を用いた車載ステレオカメラのセルフキャリブレーション

    深澤至貴, 岡谷貴之, 出口光一郎, 木村宣隆

    研究報告コンピュータビジョンとイメージメディア(CVIM) 2013 (10) 1-8 2013/01/16

    More details Close

    車載ステレオカメラのセルフキャリブレーション,つまり車両が一般道を通常走行する際に得られる画像のみを用いたキャリブレーションの方法を述べる.具体的には,各カメラの内部パラメータおよびレンズ歪がオフラインのキャリブレーションで既知であるとし,ステレオカメラを構成する左右カメラの間の相対姿勢を推定する.この場合,自然のシーンを撮影した左右カメラの画像間でいかに外れ値のない点対応を得られるか,あるいは混入した外れ値を除去できるかが鍵となる.走行中の連続画像間および左右カメラの画像間で特徴点を追跡・対応付けてその軌跡を得,それらをすべて使用して完全な3次元復元(SfM)を行い,ロバストバンドル調整によって全体最適化をすることで,得た点対応に混入する外れ値を効果的に取り除き,高精度に目的パラメータを推定する.実験結果は,この方法がステレオカメラのキャリブレーションに有効であることを示した.

  24. 車載カメラで撮影した画像列と市街地地図情報を用いた建築物の出現・消失の推定

    手塚大基, 岡谷貴之, 出口光一郎

    研究報告コンピュータビジョンとイメージメディア(CVIM) 2013 (9) 1-7 2013/01/16

    More details Close

    本研究では車載カメラで走行しながら市街地を撮影した画像群及び,建築物の輪郭情報が記載された住宅地図を使って,両者の間の不整合を自動的に特定する方法を述べる.例えば,住宅地図上には存在するが,実際には消失した建築物の自動判定が行なえる.具体的には画像群からシーンの3次元復元を行い,得られた3次元情報と地図上の建築物との比較を行い,建築物の存在を確率的に推定する.東日本大震災で被災した市街地を撮影して得た被画像と,被災前の住宅地図を使って実験を行ったので,その結果を報告する.

  25. 過去の街並みを可視化するスマートフォンを用いた拡張現実

    佐藤慎也, 岡谷貴之, 出口光一郎

    研究報告コンピュータビジョンとイメージメディア(CVIM) 2012 (3) 1-4 2012/05/16

    More details Close

    本研究では,身近なスマートフォンを用いて,過去の街並みを可視化する拡張現実(AR)を実現する.ある街の通りでスマートフォンをかざすと,その方向の過去の街並みがスマートフォンの画面上に表示され,あたかもその画面が窓となり,その中に時間をさかのぼった景色が見えるようなものである.画面に表示する画像は,あらかじめ撮影した全方位画像から切り出して生成する.このとき,端末の姿勢を内部センサおよび画像を組み合わせて推定し,リアルタイムで切り出し領域を変えて表示することで,上述の拡張現実感を得られるようにする.端末の姿勢がユーザの見たい方向と一致しない場合でも,端末の表示画面側にあるカメラを用いてユーザの顔を検出し,端末とユーザの顔の相対位置を推定し,この不一致を補正する.ユーザが今いる位置は,端末のGPSデータと,端末のカメラがその場で撮影した画像を両方利用して推定する.具体的には,GPSによっておおまかな位置を推定し,それによって全方位画像のデータベースから対象とする画像集合を絞り込む.その後,撮影画像との間で画像特徴を用いた類似度を求め,位置を高精度に推定するとともに,表示すべき画像を決定する.

  26. A Method for Recognizing Surface Qualities from Natural Images based on Learning to Rank

    2012 (7) 1-7 2012/03/08

  27. Creation of Multi-Viewpoint Panoramas for Sparse Streets with a Small Number of Buildings

    2012 (9) 1-8 2012/03/08

  28. 両眼視差と被写界深度ボケを同時に提示する注視反応ディスプレイ

    折笠達郎, 岡谷貴之, 出口光一郎

    画像の認識・理解シンポジウム(MIRU2011)論文集 2011 1201-1208 2011/07/20

  29. 画像の標本化過程を考慮した高精度かつ頑健な平面の追跡

    伊藤栄介, 岡谷貴之, 出口光一郎

    画像の認識・理解シンポジウム(MIRU2011)論文集 2011 440-447 2011/07/20

  30. 2A2-O09 Design and implementation of reflection-based tactile sensor with diffused lighting source for touch panel use(Tactile and Force Sensing (1))

    KOIKE Momotaro, SAGA Satoshi, OKATANI Takayuki, DEGUCHI Koichiro

    2011 "2A2-O09(1)"-"2A2-O09(4)" 2011/05/26

    Publisher: The Japan Society of Mechanical Engineers

    More details Close

    With the advancement of information technology device, many tactile sensing device are developing. We aim at simple but sensitive tactile sensor and have proposed reflection-based tactile sensor. The reflection-based tactile sensor can employ LED (Light Emitting Diode) and PD (Photo Diode) as thin and fast sensing devices. However this sensing method does not have optical lens. This results in low density sensing because the sensor cannot employ imaging technology. The essence of the sensor is the distribution of brightness caused by diffused light. We have investigated the changes of three dimensional distribution according to simple deformation pattern through simulation. This time we consider the capability of this method from minute simulation, and design the reflection-based tactile sensor employing LED and PD.

  31. 2P1-O14 Implementation of reconstruction method of shape of sensor surface employing total optimization on total-internal-reflection based tactile sensor(Tactile and Force Sensing(2))

    Taira Ryosuke, Saga Satoshi, Okatani Takayuki, Deguchi Koichiro

    2011 "2P1-O14(1)"-"2P1-O14(4)" 2011/05/26

    Publisher: The Japan Society of Mechanical Engineers

    More details Close

    Because of increasing demands for robots designed to work in the living environment and intuitive interfaces such as iPhones, it is necessary to develop high performance tactile sensors. We are developing a total-internal-reflection based tactile sensor, which reconstructs the sensor surface from a set of feature points on a reflection image of a image pattern. Our proposed method solves Geometric optical constraints by total optimization to reconstruct the sensor surface. By reconstruction from simulated images, we found the proposed method could perform high accurate reconstruction in real time and it was applicable for various sensor shapes and robust to tracking error. Then, we implemented the proposed method on the actual sensor and confirmed the real-time reconstruction.

  32. Hand Pose Recognition Using Markers and Appearance

    2011 (26) 1-2 2011/05/12

  33. Accurate and robust 2D visual tracking based on a model of image sampling process

    IEICE technical report 110 (381) 127-134 2011/01/20

    Publisher: 電子情報通信学会

    ISSN: 0913-5685

  34. Gaze-Reactive Display Simultaneously Presenting Binocular Disparity and Defocus Blur

    2011 (40) 1-8 2011/01/13

    Publisher: 情報処理学会

    ISSN: 1884-0930

  35. GPU based high-speed and high-precision visual tracking

    ITO EISUKE, OKATANI TAKAYUKI, DEGUCHI KOICHIRO

    2010 (30) 1-5 2010/05/20

    Publisher: 情報処理学会

    ISSN: 0919-6072

  36. Accurate Shape Measurement By Integrating Information of Stereo Triangulation and Photometric Stereo

    SEKI TAKURO, OKATANI TAKAYUKI, DEGUCHI KOICHIRO

    2010 (31) 1-7 2010/05/20

    Publisher: 情報処理学会

    ISSN: 0919-6072

  37. Pedestrian Detection Using Kernel Partial Least Squares Analysis

    ABE TAKASHI, OKATANI TAKAYUKI, DEGUCHI KOUICHIRO

    2010 (21) 1-7 2010/03/11

    Publisher: 情報処理学会

    ISSN: 0919-6072

  38. Visual recognition of wire harnesses for automated wiring

    IEICE technical report 109 (374) 59-65 2010/01/21

    Publisher: 電子情報通信学会

    ISSN: 0913-5685

  39. Visual recognition of wire harnesses for automated wiring

    IEICE technical report 109 (375) 59-65 2010/01/21

    Publisher: 電子情報通信学会

    ISSN: 0913-5685

  40. Visual recognition of wire harnesses for automated wiring

    IEICE technical report 109 (376) 59-65 2010/01/21

    Publisher: 電子情報通信学会

    ISSN: 0913-5685

  41. Visual Recognition of Wire Harnesses for Automated Wiring

    YONEDA MASAKI, OKATANI TAKAYUKI, DEGUCHI KOICHIRO

    2010 (10) 1-7 2010/01/14

    Publisher: 情報処理学会

    ISSN: 0919-6072

  42. 1A1-D07 Versatile 3D reconstruction method of reflection type tactile sensor

    Taira Ryosuke, Saga Satoshi, Okatani Takayuki, Deguchi Koichiro

    2010 "1A1-D07(1)"-"1A1-D07(4)" 2010

    Publisher: The Japan Society of Mechanical Engineers

    More details Close

    We are developing a reflection type tactile sensor, which is simple and sensitive one. This sensor measures deformation of contact surface from reflection image. In order to solve the inverse problem, we formulated new method of shape reconstruction employing three dimensional vectors. We evaluated our proposed method by reconstructing deformed surface from reflection images produced by simulation.

  43. 携帯型プロジェクタのためのプロジェクタカメラシステムの自動キャリブレーション

    岡谷貴之

    画像ラボ 21 (5) 1-6 2010/01

    Publisher: 日本工業出版

  44. Highly accurate realtime camera motion estimation using a multicamera system

    IEICE technical report 109 (182) 27-34 2009/08/31

    Publisher: 電子情報通信学会

    ISSN: 0913-5685

  45. Highly Accurate Realtime Camera Motion Estimation Using a Multicamera System

    TAKEDA GORO, OKATANI TAKAYUKI, DEGUCHI KOICHIRO

    2009 (5) 1-8 2009/08/24

    Publisher: 情報処理学会

    ISSN: 0919-6072

  46. Calibration of a Projector-Camera System for Mobile Projectors

    OKATANI TAKAYUKI, ABE DAISUKE, TAKUMA NAOHIRO, DEGUCHI KOICHIRO

    2009 (4) 1-8 2009/08/24

    Publisher: 情報処理学会

    ISSN: 0919-6072

  47. Gaze Reactive Display Simultaneously Presenting Defocus Blur and Motion Parallax

    SUZUKI TAKAAKI, OKATANI TAKAYUKI, DEGUCHI KOICHIRO

    2009 (15) 1-8 2009/06/02

    Publisher: 情報処理学会

    ISSN: 0919-6072

  48. Bundle Adjustment

    OKATANI TAKAYUKI

    2009 (37) 1-16 2009/06/02

    Publisher: 情報処理学会

    ISSN: 0919-6072

  49. A Calibration Method for a Image Projection by a Mobile Projector

    TAKUMA NAOHIRO, OKATANI TAKAYUKI, DEGUCHI KOICHIRO

    2009 (13) 1-7 2009/06/02

    Publisher: 情報処理学会

    ISSN: 0919-6072

  50. 1A1-D01 Robot Vision for Object Recognition and Self-Localization

    ABE Hiroki, SAGA Satoshi, OKATANI Takayuki, DEGUCHI Koichiro

    2009 "1A1-D01(1)"-"1A1-D01(4)" 2009/05/25

    Publisher: The Japan Society of Mechanical Engineers

    More details Close

    When a robot manipulates some objects, the object should be specified by the robot vision. This paper describes a novel object recognition approach which combines a method for generic visual categorization with three- dimensional segmentation method. The approach is as follows; first, the object area in the image is cut out by clustering 3D map of features obtained by SLAM. Next, by using the generic visual categorization method, "Bag-of-Keypoints," to the area, robot can recognize what it is and where it is in the image. Our approach contributes to the object recognition that can correspond to the diversity of objects and the image segmentation that uses 3D positions of features which compose the object. In this paper, we discuss the effectiveness of our approach through some tableware detection experiments.

  51. Appearance-based Pose Estimation Using Multiple Cameras Without Learning Process

    TAKEDA Goro, OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 2008 (36) 93-98 2008/05/01

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    Conventional appearance-based methods for estimating the pose of an object from its image use the learning samples of object appearance under known poses that are acquired in advance. These methods suffer from the increase in the time required for acquring the learning samples with the increase of the degrees of freedom of the pose parameters. Furthermore, these methods cannot deal with unknown objects other than the ones that have been learned in advance. The differential camera proposed by Yang et al., which was developed to solve these problems by using a rigid combination of a few cameras, entirely depends on the assumption that the relation between the pose change and the appearance change can be well approximated by a linear function. However, the relation is in essence highly nonlinear, which clearly leads to the limitations of the method. In this paper, we propose a method that uses a larger number of cameras than the differential camera and directly estimates the object pose without the above linearity assumption. To be specific, the method estimates the changes of the object pose by comparing the image captured at a time by a refenrece camera with the images that are captured later by other cameras and are warped by projective transformations.

  52. Realizing High-resolution Image by Superimposed Projection Using Multiple Projectors

    WADA Mikio, OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 2008 (27) 313-320 2008/03/11

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    Recently, it has been shown that it is possible to realize a higher-resolution image by superimposing multiple low-resolution images projected by different projectors. In our previous study, we pointed out several peculiar properties with the images realized by this superimposed projection based on theoretical analyses. For example, the quality of the realized images depends on the brightness structure and the overall brightness of a target image. In this paper, we show that they hold in the case of real systems by constructing a working system of superimposed projection and conducting experiments. Furthermore, we present a novel method for the calibration of the system, which is based on phase shifting that is widely used in 3D shape reconstruction. We then show that the method is superior to conventional methods in terms of calibration accuracy.

  53. Realizing High-resolution Image by Superimposed Projection Using Multiple Projectors

    WADA Mikio, OKATANI Takayuki, DEGUCHI Koichiro

    IEICE technical report 107 (539) 313-320 2008/03/10

    Publisher: The Institute of Electronics, Information and Communication Engineers

    ISSN: 0913-5685

    More details Close

    Recently, it has been shown that it is possible to realize a higher-resolution image by superimposing multiple low-resolution images projected by different projectors. In our previous study, we pointed out several peculiar properties with the images realized by this superimposed projection based on theoretical analyses. For example, the quality of the realized images depends on the brightness structure and the overall brightness of a target image. In this paper, we show that they hold in the case of real systems by constructing a working system of superimposed projection and conducting experiments. Furthermore, we present a novel method for the calibration of the system, which is based on phase shifting that is widely used in 3D shape reconstruction. We then show that the method is superior to conventional methods in terms of calibration accuracy.

  54. プロジェクタ投影像の超解像化:複数プロジェクタ投影像の重ね合わせによる高解像度画像表示

    岡谷貴之

    画像ラボ 19 (11) 65-69 2008/01

    Publisher: 日本工業出版

  55. Generation of virtual reflectance property by projector-camera system

    24 313-318 2007/10/25

    Publisher: 〔計測自動制御学会〕

    ISSN: 1343-7631

  56. Object Extraction from Images Based on Active Lighting

    OKAZAKI Tomoya, OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 2007 (42) 113-120 2007/05/15

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper presents a method for segmenting a target object from the background of a scene on its image by actively controlling lighting using projectors. The method uses the image disparity between the target object and the background scene due to the difference in their depth that emerges on the stereo pair of the camera and a projector. An identical pattern is projected onto the background scene and the scene with the target object, respectively, and two images are captured for them. Then, the image region of the target object is identified using the difference in the brightness between the two images. When there is no difference in the natural appearances of the target object and the background scene, it is impossible for the ordinary "passive" background subtraction method to segment them, whereas the proposed method can accurately segment them based on their disparity. We further present a method that uses multiple patterns to deal with the image regions such that the disparity does not result in appearance changes, and also present a method that uses multiple projectors to cope with shadows generated by the target objects. The efficacy of these methods is shown through several experiments.

  57. On Numerical Computation for Factorization of a Matrix with Missing Components

    Okatani Takayuki, Deguchi Koichiro

    CVIM 2005 (88) 123-130 2005/09/06

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper considers the problem of factorizing a matrix (or principal component analysis) with missing components. Whether a matrix is uniquely factorized or not depends on how the missing components distribute in the matrix. In this paper, we derive a sufficient and necessary condition on the missing components distribution for unique factorization. We also review the Wiberg algorithm, supposedly the most basic algorithm for the problem. We summarize the derivation of the algorithm and describe its appropriate numerical implementation, including its relation to the above uniqueness condition. We also show through experimental results that the Wiberg algorithm shows better performance than other algorithms with respect to computational cost and convergence property.

  58. カメラキャリブレーション

    植芝俊夫, 岡谷貴之, 佐藤智和

    情報処理学会コンピュータビジョンとイメージメディア研究報告 148 (18) 1-18 2005/03/03

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This article gives a survey of recent studies of geometric camera calibration techniques. After some explanations of fundamentals of projective geometry, a perspective camera model will be introduced followed by a discussion about its behavior in the three-dimensional projective space. Next various camera calibration algorithms based on this model are presented which are classified into two categories : methods using reference objects and self-calibration techniques without them. Then the refinement of estimation parameters through bundle adjustment will be explained and finally modeling and calibrating the effect of lens distortions not obeying perspective projections will be discussed.

  59. 3D Shape Recovery By Fusion of Mechanical and Image Sensors

    OKATANI Takayuki

    IPSJ SIG Notes. CVIM 2005 (4) 123-130 2005/01/21

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper presents a method that resolves several difficulties with the problems of 3D recovery from multiple images by attaching mechanical sensors, such as gyrosensors and accelerometers, to the camera and then measuring the pose of the camera from these sensors. To be specific, it is discussed how the problems of image matching. which are still difficult to perform, could be solved by introducing mechanical sensors measuring 3D pose of the camera. including discussion as to choice of mechanical sensors and their measuring accuracy. Moreover, several methods are described for calibration of the system, which is to determine the difference in internal coordinate representation between the camera and the mechanical sensors, and which will be a problem need to be carefully considered.

  60. 画像上の人物追跡のための色空間における肌色領域の分割抽出法 (第20回センシングフォーラム 資料--センシング技術の新たな展開と融合) -- (セッション1C2 人物と顔のパターン計測(2)--パターン計測部会企画)

    伊茂治 公介, 岡谷 貴之, 出口 光一郎

    センシングフォ-ラム資料 20 289-294 2003/09/16

    Publisher: 〔計測自動制御学会〕

    ISSN: 1343-7631

  61. Registration and Display of Multiple Fundus Images

    SHIMAHARA Tatsuya, NAKAJIMA Taira, OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 2003 (41) 31-36 2003/05/08

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper presents a method for registering fundus images. The fundus image is the photograph of inner bottom of an eyeball. Because it is taken through the pupil and the view field is restricted to small region, it is useful to register multiple fundus images in diagnosing. In this paper we register the fundus images employing the Ali Can's algorithm. This algorithm has the advantage of registering automatically and accurately when the overlap is wide. But when the region of overlap is narrow, we often fail because of the lack of the correspondences. We discuss the increase of the member of correspondences when the overlap is narrow and propose a modified method to obtain more accurate registration.

  62. On Tracking Multiple Objects Using Mean Shift Algorithm

    KAWANAKA Daiki, NAKAJIMA Taira, OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Technical Report CVIM 2003 (41) 9-14 2003/05/08

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper presents a method for tracking mulitple objects in an image sequence in a real time. If the objects can be accurately extracted from the images by background subtraction, it is possible to robustly track the objects with a small amount of computational complexity using the mean shift algorithm. When there are mulitple moving objects in the images and they frequently overlap with each other, it is necessary to identify each of them correctly to track them. An algorithm for this problem is shown which is based on mean shift algorithm with probabilistic weights that are computed from color distribution of each target. The weights are generated at every time when occlusion occurs and they are computed from comparison between the histogram of the target window and those of the overlapped windows. The histograms are periodically updated when there is no occlusion.

  63. Calibration of Multi - View Cameras for 3D Scene Understanding

    DEGUCHI Koichiro, OKATANI Takayuki

    IPSJ SIG Notes. CVIM 2002 (2) 1-8 2002/01/17

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    Abstract : To understand a dynamic 3D scene by multiple cameras, the characteristics as well as locations and poses of the cameras must be identified. For such the dynamic scene understandings, the conventional camera calibration techniques using some specific calibration objects has a limitation. In the application here, the object shapes are unknown, and some camera parameters such as focal length and camera- pose like to be variable. Then, so-called the self-calibration must be employed, where the camera parameters and the object shapes are identified simultaneously and dynamically. In this paper, we introduce our strategy of implementation for the calibration for multi-view cameras based on the self-calibration, show the experimental results, and discuss the problems concerning the dynamic calibration of the multi-view cameras.

  64. 画像からのカメラの姿勢 3姿勢形状復元における推定精度の限度の限界について

    岡谷貴之

    第6回画像の認識 理解シンポジウム講演論文集 335-340 2002

  65. Appearance-based Pose Estimation from Images by Fitting Linear Models

    DEGUCHI Koichiro, OKATANI Takayuki

    18 19-24 2001/10/03

    Publisher: 〔計測自動制御学会〕

    ISSN: 1343-7631

  66. Estimating camera translation based on a voting method using a camera with a 3D orientation sensor

    OKATANI Takayuki, DEGUCHI Koichiro

    2001 (87) 25-32 2001/09/13

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper is concerned with the problem of estimating pose of a camera in space from a pair of images of a scene or an object that is taken with the camera in a case where the point correspondences between the images are unknown. The camera orientation is obtained by a 3D orientation sensor rigidly attached to the camera. Assuming that the intrinsic parameters of the camera are known, we estimate only the camera position, i.e. translation between viewpoints, from the images and the camera orientation obtained from the sensor. This camera translation is estimated by voting pairs of points that likely correspond to each other. This is done without explicitly determining the point correspondences. Several methods using random samples have been applied to these kinds of problems and their effectiveness has been already known. We compare these methods with our voting-based method in terms of robustness.

  67. Why does the eigen space method work well?

    DEGUCHI Koichiro, OKATANI Takayuki

    IPSJ SIG Notes. CVIM 2001 (66) 1-8 2001/07/05

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    The eigen psace method has established its position among computer vision techniques, recently, as a tool for robust image analysis as well as efficient image data compression method. However, image data itself has a significant difference from statistical data to which the conventional eigen space method is conceptually applicable, because its dimensionality is much larger than the number of samples which are used to construct the eigen space. Here, we claim that the reduction of the dimensionality which is carried out in the process of the eigen space method plays an important role for model fitting to the image data. In this article, we introduce the image based pose estimation task as an example to apply the eigen space method. We make comparison between the eigen psace method and the direct interpretation method where a linear model combining poses and images. Then, we discuss the comparison from the viewpoint of the theory of over-learning.

  68. A System for Real - time 3D Reconstruction from Multiple Motion Images

    TAKEDA Jun, OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 2001 (36) 75-80 2001/05/10

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This article reports a configuration of a system for real-time 3D reconstruction from multiple motion images for 3D motion capturing. We employ the volume intersection(or, so-called the shape from silhouette)technique. In this system, multiple images are obtained by cameras arranged around a moving object. From the respective images back-ground images are subtracted, then, the object images are back-projected into 3D space. Finally, the volume intersection of each projection cones is extracted. This technique has great advantages of the simple and high-speed processing. Our system has 6 digital camera and one PC, and reconstruct 3D object in(180cm)^3 space with 1cm resolution at 3 or 4 frames/s speed. It is possible to speed-up to over 10 frames/s when the resolution is reduced to 2cm. We report the techniques to achieve the processing capability as well as the total configuration of the system.

  69. Multiple Image Maching for 3D Object Reconstruction

    HITOMI Yasunobu, TAKAHASHI Katsuhiko, OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 2001 (36) 81-86 2001/05/10

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This article reports a technique for extracting correspondences from multiple images to reconstruct 3D shape of the object. The images are obtained by uncalibrated cameras, but assumed to have 6 or more points whose world coordinates are known. From those points, first, relative poses and positions of the cameras are determined. Then, the 3D positions of the other points are estimated based on the stereo principle. To establish these correspondences, we employ the technique proposed by Zhang(1994). But, in his method, the process called as the relaxation was not clearly defined and did not efficiently exclude fault correspondences. We modified this process and achieved an effient and accurate matching. We show total process of image matching based on the epipolar geometry, and also show the results of 3D reconstruction of large buildings from a set of aerial images.

  70. Estimation of Convexity/Concavity of Smooth Surface from Shading Using Distribution of Isophotes

    OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 1998 (5) 49-55 1998/01/22

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper is concerned with the problem of identifying types of image singular point as maxima, minima, or saddle points, especially for the case of unknown lighting direction. (Singular points are defined as maximally bright points in the image whose corresponding surface normals coincide with the lighting direction). The identification of singular points is the key to the global shape from shading problem, and it is equivalent to determine convexity/concavity of the object shape. Using photometric invariance that saddle points of the image grey level profile always lie on the parabolic curves (curves of zero Gaussian curvature), it is shown that the two singular points which are connected by two steepest ascent curves starting from an image saddle point have different signs with each other. This constraints the combination of the types of the singular points in the global shape from shading problem.

  71. Global Shape from Shading : A Method for Classifying Singular Points Using Isophotes

    OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 1997 (70) 187-194 1997/07/24

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    It has been shown that the shape of a smooth object can be globally reconstructed from one shaded image if singular points in the image are correctly classified into minima, maxima and saddle points. This paper shows that the topology of isophotes (lines of equal brightness) around the singular points puts some constraints on the classification. They come from the relation in terms of differential geometry between a smooth surface and its shaded image. In some cases, the numerical solution of a PDE (the image irradiance equation) that the conventional methods required is no longer necessary. This makes it possible to accurately classify singular points and to reliably reconstruct the surface shape.

  72. Segmentation of Surface Curvature Using Multiple Images Acquired under Different Lighting Directions

    OKATANI Takayuki, DEGUCHI Koichiro

    IPSJ SIG Notes. CVIM 1997 (31) 15-22 1997/03/27

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper describes a new method for determining the sign of Gaussian curvature at points on the object surface from three images acquired under different lighting directions. The determination can be done without knowing the directions and the strength of the light sources and also surface albedo. The basic idea is well explained through the relationship between the Gaussian image of the surface and the image brightness. The proposed method doesn't need any special condition for the light source directions which was assumed in the conventional methods. It is shown that the proposed method produces good results even in the case with non-Lambertian reflectance.

  73. Reconstructing Shape from Shading with a Point Light Source at the Projection Center Shape Reconstruction from an Endoscopic Image

    Okatani Takayuki, Deguchi Koichiro

    IPSJ SIG Notes 1996 (5) 19-26 1996/01/18

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper describes an approach to reconstructing a shape from its shaded image in the case where a point light source is located at the projection center. This condition well approximates the imaging system of an endocope. In this case, the image gray level depends on not only the gradient but also the distance from the light source of each point on the surface. To deal with this difficulty, we introduce the evolution equation for equal-range contours on the surface. Propagating this contour by solving the equation, we can reconstruct a shape. Experimental results for real medical images show feasibility of this method.

  74. Reconstructing Surface Shape from Shading Using Eigenspace Method

    Okatani Takayuki, Deguchi Koichiro

    IPSJ SIG Notes 1995 (33) 1-7 1995/03/23

    Publisher: Information Processing Society of Japan (IPSJ)

    ISSN: 0919-6072

    More details Close

    This paper describes a new approach to reconstructing surface shape from its shading. By assuming that a given surface consists of a set of locally quadric segments which belong within a specific class, the surface shape is determined excepting its sign of convexity. The local segment is estimated by using the correspondence between quadric surface parameters and its shading. The correspondence is calculated prior to the estimation and stored. If this correspondence is prepared for every raw image shading, very large strage space is required. We use eigenspace method for the reduction of data size.

Show all ︎Show first 5

Books and Other Publications 7

  1. 深層学習

    岡谷, 貴之

    講談社 2022/01

    ISBN: 9784065133323

  2. 質感の科学 ―知覚・認知メカニズムと分析・表現の技術―

    小松英彦, 岡谷貴之

    朝倉書店 2016/10/20

  3. 深層学習

    麻生 英樹, 安田 宗樹, 前田 新一, 岡野原 大輔, 岡谷 貴之, 久保 陽太郎, ボレガラ ダヌシカ

    近代科学社 2015/11/05

  4. 深層学習(機械学習プロフェッショナルシリーズ)

    岡谷貴之

    講談社 2015/04/07

  5. コンピュータ工学入門

    鏡 慎吾, 佐野 健太郎, 滝沢 寛之, 岡谷 貴之, 小林 広明

    コロナ社 2015/03/01

  6. コンピュータビジョン最先端ガイド6 (CVIMチュートリアルシリーズ)

    岡谷貴之

    2013/01/01

  7. コンピュータビジョン最先端ガイド-CVIMチュートリアルシリーズ-

    岡谷貴之, 八木康史, 斎藤英雄

    2011/04

Show all Show first 5

Research Projects 26

  1. 専門知に基づいて画像を理解し説明する対話型AIの実現

    岡谷 貴之, 菅沼 雅徳

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業

    Category: 基盤研究(A)

    Institution: 東北大学

    2023/04/01 - 2027/03/31

    More details Close

    研究実施計画に従って研究を進め、以下のような成果を得た。 まず、車載カメラの画像から運転時の危険を予測・説明するタスクを設計し、そのためのデータセットDHPR(Driving Hazard Prediction and Reasoning)を作った。既存の車載画像のデータセットの画像に,クラウドソーシングでアノテーションを行った。その結果をまとめたものは現在、論文誌に投稿中である。 研究項目「画像記述表現高度化」では、マルチモーダルAIのための画像特徴抽出の方法を研究した。さらに、画像からの異常検知を題材に、画質記述表現の高度化につながる特徴抽出に関する研究を複数行い、国際会議INDIN2023、 WACV2024等で発表した。また、画像セグメンテーションのための無教師ドメイン適応手法を新たに開発し、Computer Vision and Image Understanding誌にて発表した。 研究項目「知識の表現・利用方法」では、上述の画像特徴抽出方法を大規模言語モデルと統合したマルチモーダルAIモデルを構築した。上述のDHPRを用いて、複数のモデルの学習(文脈内学習含む)と推論性能の評価実験を行い、一定の精度で推論を行えることと、実用レベルまでには改善の余地が残ることを確認した。また、橋梁の画像点検タスクを対象に、橋梁の変状を認識し説明可能なマルチモーダルAIのモデル構築と性能評価を行った。成果はComputer-Aided Civil and Infrastructure Engineering誌にて発表した。さらに、未知の屋内環境を探索し、地図を構築するタスクを対象に、暗黙知を学習し推論に活用できるAIモデルの研究を行い、International Journal of Computer Vision誌にて発表した。

  2. 深奥質感のマルチモーダル深層モデルの確立

    岡谷 貴之, 菅沼 雅徳, 鈴木 潤, 劉 星

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業 学術変革領域研究(A)

    Category: 学術変革領域研究(A)

    Institution: 東北大学

    2020/11/19 - 2025/03/31

    More details Close

    まず,DNNによる画像理解の研究を行い,複数の成果を得た.第一に,自然言語の指示によって作業を行うAIエージェントを開発した.エージェントが空間を見る視野をなるべく広く取るとともに,作業の指示を与える言語情報を2回解釈することに特徴があり,これによって高い作業成功率の達成が可能となった.手法は,国際的なベンチマークテストAlfredにおいて論文発表時点で全世界トップの性能を達成するとともに,成果の一端を国際会議IJACI2021にて発表した. 第二に,画像理解タスクの性能向上を目指して,画像記述の研究を行った.まず,これまでの手法が「比喩的表現」を基本的に行えないことに着目し,データセットとベンチマークテストを設計するとともに,評価方法を検討した.成果の一部は国際会議に投稿中である.第三に,画像記述のためのより高性能かつ高効率なDNNの設計を行った.画像特徴を取り出すのに,従来手法のように物体検出器に由来するCNNを使うのではなく,トランスフォーマーを用いることで高速化を達成した.同時に,物体領域から抽出した特徴と,画像を格子状に分割した領域から取り出した特徴の双方を効果的に統合することで,記述の精度を大幅に向上させている.国際的なベンチマークテストCOCO captionsにおいて,世界トップクラスの記述精度を従来手法の少なくとも数分の1の計算量で達成可能なDNNとなっている.本成果は国際会議に投稿中である.さらに,主に自然言語で表現された外部知識を,画像理解に導入する手法の検討を行っており,継続中である. また,以上とは独立に,自己教師学習の方法について研究を行った.特に近年活発に研究されている負例を用いない画像特徴の自己教師学習手法について,その有効性がどこから由来するものかを理論的に検討した.成果は国際会議に投稿中である.

  3. 深層学習がもたらすパラダイムシフトを加速するネットワーク設計理論の構築

    岡谷 貴之, 菅沼 雅徳

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業 基盤研究(A)

    Category: 基盤研究(A)

    Institution: 東北大学

    2019/04/01 - 2023/03/31

    More details Close

    当初計画に従い3つの項目について研究を進めた. 1つ目は画質改善タスクをターゲットとした深層ネットワークの構造の研究である.タスクを選ばないユニバーサルなネットワーク構造存在するのか,どのようにすれば実現できるかを検討したところ,一定の成果を得た.ただし訓練データと同じドメインのデータに対しては良好な性能を見せるものの,ドメインが多少なりとも変化すると性能が急激に低下する傾向があることがわかり,計画を考え直すに至った.ただしここまでの成果はarXiv上に論文として発表している. 2つ目は推論の不確かさを予測する方法である.これについては,クラス分類にコサイン類似度を用いたロジスティック回帰を行う構造をネットワークに持たせ,学習を行うことで,学習外分布(out-of-distribution)の入力を高精度に検出できることを実験的に示した.その性能は世界最高水準にあ流.以上の成果は国際会議ACCVにて発表した.この他にも,学習済みの分布と学習外分布を完全に分離して捉えるのではなく,その中間に位置するサンプルを考慮し,学習に使うことで,学習外分布の入力をより高精度に検出できることを示した.中間に位置するサンプルは,学習画像の画質を低下させる画像処理によって作られた画像を用いる.この方法は,上述の方法よりさらに高精度に,分布外入力の検出が行えることを実験を通じて示した.成果は現在国際会議に投稿中である. 3つ目は当初構想していた方法がうまく進まなかったため,方針を変更し,その検討で得た問題意識に従い,画像からひびを検出する方法の精度限界がどこにあるかを調べることを行った.訓練データ内のひびのアノテーションのエラーがひびの検出にどのように影響するかを,さまざまな角度から実験によって調べ,いくつかの知見を得た.成果は国際会議に投稿中である.

  4. Deep Representation Learning of Shitsukan Using Vision and Language

    Okatani Takayuki

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research on Innovative Areas (Research in a proposed research area)

    Category: Grant-in-Aid for Scientific Research on Innovative Areas (Research in a proposed research area)

    Institution: Tohoku University

    2015/06/29 - 2020/03/31

    More details Close

    We have obtained step-by-step achievements towards realizing an AI that can recognize the Shitsukan of an object from its single image. We have developed the following methods: i) a CNN-based method for Shitsukan recognition based on learning to rank image pairs; ii) a method for mining visually recognizable Shitsukan concepts from pairs of images and their descriptions on the Web; iii) a method for generating appropriate titles for images of products while capturing the context-aware Shitsukan concepts. The developments of the above lead to the conclusion that Shitsukan recognition is "comprehensive image recognition," and we successfully developed a network with an attention mechanism, which is an offshoot of today's multimodal deep neural networks for image understanding.

  5. Supporting performance-aware programming with machine learning techniques

    Hiroyuki Takizawa, Kobayashi Hiroaki, Suda Reiji, Okatani Takayuki, Egawa Ryusuke, Ohshima Satoshi

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)

    Category: Grant-in-Aid for Scientific Research (B)

    Institution: Tohoku University

    2016/04/01 - 2019/03/31

    More details Close

    This work has demonstrated some case studies of effectively using machine learning techniques for supporting High-Performance Computing (HPC) programming. Various problems in code optimization can be solved by converting the problems to the problems that have already been proven to be solved by machine learning. Moreover, this work clarified the importance of analyzing the target problems in advance of machine learning, because it is unlikely that a sufficient number of training data are available in code optimization problems. Moreover, as well as HPC programming, machine learning also needs knowledge and experiences of human experts. However, in machine learning, the problem is already parameterized, and hence can be solved if sufficiently-high performance is available.

  6. Spatio-Temporal City Modeling for Visualizing Damages of and Recoveries and Reconstruction from a City-scale Disaster

    Okatani Takayuki

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)

    Category: Grant-in-Aid for Scientific Research (B)

    Institution: Tohoku University

    2013/04/01 - 2016/03/31

    More details Close

    We have studied methods for spatio-temporal modeling of a city from image sequences of its streets acquired periodically by a vehicular camera, having developed a number of methods each of which can be selected and used depending on the purpose and the types of available data. Meanwhile, targeting at the disaster areas of the Greate East Japan Earthquake, we have periodically run a vehicle with a camera and have created an image archive recording the recovering processes of the disaster area, the amount of which is 40 terabytes including the data we had captured before this project. Additionally, the performance of the developed methods were examined using the same data. These achievements have paved the way to be able to visualize damages right after the incident, monthly recovery process, and yearly reconstruction process seen in the disaster area as well as to quickly grasp damages of future natural disasters.

  7. ものの質感を画像を元に認識するシステムの実現

    岡谷 貴之

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業 新学術領域研究(研究領域提案型)

    Category: 新学術領域研究(研究領域提案型)

    Institution: 東北大学

    2013/04/01 - 2015/03/31

    More details Close

    本年度は,質感認識の新しい手法を実現すべく,ディープラーニングを用いて質感および物体の材質カテゴリーを認識する方法を中心に,以下に述べるような研究を行った. まず,前年度構築したシステムと学習用データセットを利用して,多層畳込みニューラルネットの訓練を行い,結果を評価した.質感認識については,13の質感を表す形容詞に関する画像ペア間の順序情報を,各形容詞につき1000件程度,被験者を集めて収集し,新しいデータセットを構築した.以前われわれは,ニューラルネットを用いない方法で画像から取り出した特徴を,サポートベクトルマシンでランキングする方法を提案していた.本計画で新たに,物体認識用に訓練された畳込みニューラルネットを使って,それが画像から取り出す特徴量を代わりに用いて同様の実験を行った.その結果は,われわれが以前得ていた結果を相当上回り,ペアの順序の予測精度で10%程度向上し,トータルで90%前後の質感の認識性能を達成することが分かった. 次に,ディープ畳込みニューラルネットワークをマテリアル認識へ応用し,特に既存のデータセットFlickr Material Databse(FMD)での認識精度向上を図った.具体的には,ネットワークのモデルの検討と,転移学習やデータ拡張などの様々な手法の組み合わせを検討し,性能評価実験を繰り返し行なった.その結果,単一モデルで75%程度の認識精度を達成できた. さらに,質感認識の核でもある反射成分の物理的な分離手法の検討を行った.非負値行列分解を用いることで,従来法よりも高精度に分離が行えることを示した.同成果は国際会議ACCV2014にて発表し,Best Student Paper賞を受賞した.

  8. 画像のボケと両眼視差を同時に再現する注視反応ディスプレイ

    岡谷 貴之

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業 新学術領域研究(研究領域提案型)

    Category: 新学術領域研究(研究領域提案型)

    Institution: 東北大学

    2011/04/01 - 2013/03/31

    More details Close

    研究計画開始時,両眼視差とDOFボケを同時提示可能なシステムを構築し,それを用いてDOFボケが質感知覚に与える影響を調べることを研究の主要な目的としたが,初年度の23年度中には同システムの構築を終え,さらに実験もかなりの部分を終えることができたので,今年度は,さらなる詳細な検証実験の実施と,さらにこれまでの研究で見えてきた新しい課題に取り組んだ.さらに,それら成果を踏まえて,本領域の計画班や他の公募班と共同研究を実施,あるいはその準備も行った.まず取り組んだ課題は,23年度実績報告書記載の通り,質感属性の画像認識アルゴリズムの研究である.研究の結果,従来の研究に比べて,自然画像を扱えるようになり,透明感や滑らかさ,暖かさといった従来扱いにくかった質感属性を対象とできるようになった.この成果は国際会議ICPRにて公表し,さらにその内容を発展させたものを専門誌に投稿準備中である.次に,同新学術領域に参加する他の公募班(新潟大学,電気通信大学,大阪大学,京都大学の各研究グループ)と,共同研究を実施した.現在も研究中の課題を含むため詳細は省くが,本研究計画で実現した注視計測システムの生体計測への応用,われわれが実現した質感認識手法と言語的特性に基づく質感認識手法との融合による新しいアプリケーションの実現,同手法と生体の質感認識方法との比較などを含む.これらの成果は近いうちに公表される予定である.

  9. ボケを用いた立体ディスプレイシステムの研究 Competitive

    2011/04 - 2013/03

  10. Measuring States of Plant Photosynthesis by Structured Light Projection

    OKATANI Takayuki

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Challenging Exploratory Research

    Category: Grant-in-Aid for Challenging Exploratory Research

    Institution: Tohoku University

    2011 - 2013

    More details Close

    We have studied the application of a range of computer vision techniques of separating reflection components to improving measuring accuracy of chlorophyll fluorescence. One of the main contributions is that we have confirmed that the method of projecting structured light to plants and separating the direct and indirect (or global) components can really extract sub-surface scattering of incident light onto leaves of plants, resulting in the improvement of measuring chlorophyll fluorescence. We have also developed a new method of separating specular and diffuse reflection components based on the dichromatic reflection model.

  11. Study of Super Accurate Methods for Geometric Inference from Images

    OKATANI Takayuki

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)

    Category: Grant-in-Aid for Scientific Research (B)

    Institution: Tohoku University

    2010 - 2012

    More details Close

    This study is concerned with the problems of estimating geometric information such as the three dimensional shape of an object from its image(s) captured by a camera. We have sought for the theoretically most accurate methods along with development of their applications. Our achievements include a new numerical method for matrix factorization, a calibration method for projector-camera systems, a super-accurate planar tracking method, a method for spatiotemporal modeling and visualization of a city from its images captured by a vehicle-mounted camera, a calibration method for vehicle-mounted stereo cameras, a super-accurate shape measuring method based on the integration with photometric methods, new optimization methods for Markov random fields, etc.

  12. Establishing the Information Theoretical Approach to Active Sensing and Recognition

    DEGUCHI Koichiro, OKATANI Takayuki, SAGA Satoshi

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)

    Category: Grant-in-Aid for Scientific Research (B)

    Institution: Tohoku University

    2009 - 2011

    More details Close

    We achieved the main object of this research project to establish a methodology of information theoretical approach for active sensing systems. With this achievement, we have also established an analysis of mechanisms of control in human visual sensing activities. We considered how our human sensors work out their strategy to obtain most effective information. The main key was the"mutual information", which estimates the amount of information flow between an information source point and its destination. Based on this notion, we accomplished a new framework of description for active sensing systems.

  13. 複数プロジェクタによる高精細映像投影システムの研究 Competitive

    2007/04 - 2008/03

  14. 複数のカメラやプロジェクタからなる映像システムのキャリブレーション手法の研究 Competitive

    2000/04 - 2008/03

  15. Study of Multi-projector Displays Realizing Easy Projection of Super-high-resolution images

    OKATANI Takayuki

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Young Scientists (A)

    Category: Grant-in-Aid for Young Scientists (A)

    Institution: Tohoku University

    2007 - 2008

  16. Development of Active Visual Sensing Mechanism for Realizing Fast and Stable Robot Motions

    DEGUCHI Koichiro, OKATANI Takayuki, USHIDA Shun, SAGA Satoshi

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)

    Category: Grant-in-Aid for Scientific Research (B)

    Institution: Tohoku University

    2006 - 2008

  17. 自己校正理論に基づくプロジェクタカメラシステムの研究

    岡谷 貴之

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業 若手研究(B)

    Category: 若手研究(B)

    Institution: 東北大学

    2005 - 2006

    More details Close

    複数台の映像プロジェクタを組み合わせて大画面高精細画像を投影するシステムがある.個々のプロジェクタの解像度を台数分合計した,高解像度を実現できるため,今後の利用が期待されている.この種のシステムにおいて最も重要となるのが,最終的に生成される投影像の画質を左右する,システムのキャリブレーションである.このキャリブレーションには,2種類あり,一つが複数の異なるプロジェクタが投影する,個々の画像同士がきれいにつながるようにするための「位置合わせ」であり,もう一つが,そのようにつなぎ合わされた画像が正しいアスペクト比の正確な長方形となるようにするための「正規化」である.キャリブレーションには通常カメラを別途用い,カメラで撮影した画像が利用される.従来は,キャリブレーションの2要素のうち,「位置合わせ」についてはカメラを使った方法で実現されていたが,「正規化」は,投影先となるスクリーンの形の情報や,あるいはプロジェクタに別途取り付けた角度センサなど,カメラの画像以外の事前知識や他の情報を必要としていた・本研究では,位置合わせに加えて正規化についても,カメラで撮影した画像のみで行えることを理論的に証明するとともに,そのための具体的なアルゴリズムを導いた.これにより,プロジェクタをただ雑然と並べた後,キャリブレーション用のパタンを各プロジェクタから投影させた後,それを手持ちのカメラで一枚画像を撮影すれば,その画像のみから上述の2種類のキャリブレーションを同時に,完全に実行できるようになる.このことは,複数の映像プロジェクタによって大画面高精細画像を生成するシステムの利便性を大幅に高める.この結果,通常のフラットパネルディスプレイでは実現が難しいような,超高精細の画像を,身近なものとすることが期待される.

  18. Research for 3D Space Map Acquisition for Robot by Combining Active Motion and Visual Perception

    DEGUCHI Koichiro, OKATANI Takayuki, USHIDA Shun

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)

    Category: Grant-in-Aid for Scientific Research (B)

    Institution: Tohoku University

    2003 - 2005

    More details Close

    This research project aimed to develop an active vision strategy to let a robot obtain 3D scene understanding from his stereo eyes. That is, the robot constructs a 3D map of an unknown environment by himself using his own eyes (cameras) and his own intentional actions. The task to get to know the external world using vision is absolutely difficult for artificial systems, such as robots, while our human achieves it easily without teacher or advance knowledge. In this research, we placed an answer based on a key concept of the invariance against our actions. We reported a realization of the direct perception of three-dimensional space in a robot. We proposed a method to calibrate the stereo eyes of a robot to build 3D map in his brain. It does not need an external reference of calibration objects but a combination of active motion and visual perception. The idea is common to so-called the self-calibration to some extent. But, the calibration and the following construction of the 3D map still considered separately in the self-calibration. In our method, we combine action and vision to achieve totally efficient calibration and the construction of the 3D map. We rely only on the consistency in the calibration results. We introduced the fact that "stationary object both in the real environment and on the robot's 3D map never moves even when the robot moves around". This means that we accept a calibration error and resulting geometric distortion of the constructed 3D map. We consider that the most important for the 3D map is the consistency between the intended motion of the robot and the perceived image by the robot. We showed the feasibility of this idea with plenty of simulation experiments and implementation on actual robots.

  19. 能動視覚による動的な空間知覚と立体形状認識機構の解明とその応用システムの構築

    出口 光一郎, 中島 平, 岡谷 貴之

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業 特定領域研究

    Category: 特定領域研究

    Institution: 東北大学

    2003 - 2003

    More details Close

    本研究では、人間の視覚の運動機構に関する知見と、研究代表者らのこれまでの能動視覚に関する研究成果を実用化に結びつけるべく、統一的な能動視覚の原理を確立するために、平成13年度に開始した特定領域研究課題を継続・発展した。その中で、本年度は特に次の3つのシステムの実現をはかった。 (1)能動カメラによる追跡に基づく高速移動体の実時間形状認識システム (2)協調する複数カメラからの画像の統合による移動体、静止物体の実時間形状認識システム (3)能動的にカメラを移動させながら撮影した多方向からの画像による高精度形状認識システム 第一のシステムでは、たとえば、移動ロボットの前方を横切る物体、監視カメラの前方を横切る車両などの立体形状を実時間で認識する。このとき、シーン中を移動する物体を発見したならば、それに視点を固定して追跡するようにカメラのパン、チルトを制御することで、立体形状が容易に計測しうる数理的な仕掛けが、平成13年度の研究よって示されている。本年度では、この原理をシステム化し実用性を確認した。 第二のシステムでは、人間の両眼を模した2台のカメラが注視点をそろえて協調して対象の動きをとらえ、形状を計測するシステムを構築した。 その結果を、多数(10台程度)のカメラによるシステムへ拡張し、いわゆる、モーションキャプチャの新しい方式を提案し、実システムとして提示する。 第三のシステムでは、自由にカメラを移動させながら対象を撮影し、得られた画像系列から対象の立体形状を再現する。画像上の対応点の動きからカメラの軌道を再現し、対象形状を復元する。対象の形状に応じたカメラの効率的な移動方法があるか、それは、どのような要因で決定されるのかなどを理論化することも、研究の目的であった。 そして、これらの成果を学術雑誌、国際会議、国内学会にて公表した。

  20. Study on Methods for Reconstruction of 3D Model from Images Competitive

    1999/04 - 2002/03

  21. 能動視覚による動的な空間知覚と立体形状認識機構の解明とその応用システムの構築

    出口 光一郎, 中島 平, 岡谷 貴之

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業 特定領域研究

    Category: 特定領域研究

    Institution: 東北大学

    2002 - 2002

    More details Close

    本研究では、人間の視覚情報処理における眼球の運動制御を含めた能動的な視覚の仕掛けを解明するとともに,能動視覚をもつロボットを設計して,その視覚による空間知覚と運動制御を実現するシステムを構築する。 本年度は、このための1つのシステムの実現である,「行動と画像変化を結びつけたロボットの空間認識」について研究した. 人間は,2眼を用いることにより環境中の物体間の距離あるいは自分自身と物体までの距離を知ることができる.ただし,正確な絶対距離や,正確な方向を認識しているわけではない.対象の空間的な位置についての情報の幾らかは,静止している観察者によっても抽出し得るが正確ではない.しかし、観察者が動き始めれば,さらに多くの情報が手に入るようになる.移動を行うことは,それがうまく実行されるために多くの空間の情報を必要とする.正確な空間情報のもとで正確な行動をすると,眼から得られる画像は予想通りの変化をする.一方,この空間情報が不十分であると正確な行動ができないが,それに加え,不自然な画像変化をも引き起こす.この不自然な画像変化は,それによってまた移動したその観察者に多くの情報をもたらし得る.本研究では,このような行動と画像変化の結びつきに注目した,以下の研究を行った。 1.能動カメラによる追跡に基づく高速移動体の実時間形状認識システム 2.協調する複数カメラからの画像の統合による移動体、静止物体の実時間形状認識システム 3.ロボットに搭載した能動的にカメラによる、ロボットの空間認識 そして、これらの成果を統合し、人間の視覚情報処理における眼球の能動制御の仕掛けを解明するとともに、能動視覚をもつロボットを実現して、その視覚による空間知覚と運動制御を実現を図った。

  22. Study of 3D Fundus Pattern Reconstruction and Display from Fundus Images

    DEGUCHI Koichiro, NAKAJIMA Taira, HONTANI Hidekata, OKATANI Takayuki

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)

    Category: Grant-in-Aid for Scientific Research (B)

    Institution: Tohoku University

    2000 - 2002

    More details Close

    In this research, we develop a method to reconstruct and display 3D Findus Pattern from a set of multiple Findus Images taken in many directions. The fundus is the inner bottom of an eyeball. It is the only place in our bodies where blood vessels are able to be observed directly, and it is very useful part for a variety of diagnosis on helth conditions. However, it is difficult to locate the 3D positions of the fundus parts position and measure their dimensions, because the fundus observation is made through eye lens. Due to this, the diagnosis is tend to be emphasized on the subjectivity of the viewers. Some methods have been proposed where, after camera calibrations, 3D fundus display was obtained from a pair of stereo images. But it provided only a narrow area of 3D fundus because of the limited view angle of the fundus camera. In this research, we proposed the method of 3D Reconstruction of large area and showed the experimental results for some real fundus images. Findus images are skewed due to optical systems such camera lens and eye lens. The camera optical system can be fully calibrated, while the human eye system is difficult to be calibrated. In the proposed system, the optical system of the eye is modeled with a single lens, and the shape of the fundus is assumed to be sphere. It is proven that the real images of fundus through a single lens form quadratic surfaces. The deformation of fundus images reflects the parameters of the quadratic surfaces. So, from the multiple fundus images, the parameters were identified, and through which the skews were estimated. Then the 3D display in large area was achieved. To put our result to practical use, it still needs some doctors' help. But many experimental results promise the usefulness of the developed method. This is also a development of new area in the field of computer vision.

  23. 能動視覚による動的な空間知覚と立体形状認識機構の解明とその応用システムの構築

    出口 光一郎, 中島 平, 岡谷 貴之

    Offer Organization: 日本学術振興会

    System: 科学研究費助成事業 特定領域研究(C)

    Category: 特定領域研究(C)

    Institution: 東北大学

    2001 - 2001

    More details Close

    人間の能動的な視覚機構の数理的な構造、特に注視のための身体運動と眼球運動の視覚情報処理における役割と、立体形状知覚センサーとしての仕掛けを解明を目的として、その機構を模した自らも高速で移動したり首を振ったりできる能動的な搭載カメラによって、刻々変化する空間環境をとらえ、カメラの置かれている空間や対象の形状を実時間で認識するシステムを構築した。さらに、このとき、このような目的で認識にとってよりよい画像を得るような運動をカメラ自らが想起して行動するための、能動視覚の基本原理を確立した。 本年度は、特に次の3つのシステムの構築を目指し、一部を実現した。 (1)能動カメラによる追跡に基づく高速移動体の実時間形状認識システム-たとえば、移動ロボットの前方を横切る物体、監視カメラの前方を横切る車両などの立体形状を実時間で認識する (2)協調する複数カメラからの画像の統合による移動体、静止物体の実時間形状認識システム-人間の両眼を模した2台のカメラが注視点をそろえて協調して対象の動きをとらえ、形状を計測する (3)能動的にカメラを移動させながら撮影した多方向からの画像による高精度形状認識システム-自由にカメラを移動させながら対象を撮影し、得られた画像系列から対象の立体形状を再現する そして、これらの成果を統合し、人間の視覚情報処理における眼球の能動制御の仕掛け、能動視覚をもつロボットにおけるその視覚による空間知覚と運動制御を考察した。

  24. Study of Visual Servoing Robot Control without High Level Image Processing Technique

    DEGUCHI Koichiro, OKATANI Takayuki, HONTANI Hidekata

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B).

    Category: Grant-in-Aid for Scientific Research (B).

    Institution: Tohoku University

    1998 - 2000

    More details Close

    In this research, a general scheme to represent the relation between dynamic images and camera and/or object motions is developed for applications to visual control of robots. We consider the case where a moving camera observes moving objects in a static scene. The camera first obtains images of the objects moving within the scene. Then, the possible combinations of the camera and the objects' poses and the obtained images are not arbitrary but constrained to each other. Here we represent this constraint as a lower dimensional hyper surface in the product space of the whole combination of their motion control parameters and image data. The visual control is interpreted as to find a path on this surface leading to their poses where a given goal image will be obtained. In this paper, we propose a visual control method to utilize tangential properties of this surface. First, we represent images with a composition of a small number of "eigen images" by using K-L (Karhunen-Loeve) expansion. Then, we consider to reconstruct the eigen space (the eigen image space) to achieve efficient and straightforward controls. Such reconstruction of the space results in the constraint surface being mostly flat within the eigen space. By this method, visual control of robots in a complex configuration is achieved without image processing to extract and correspond image features in dynamic images. The method also does not need camera or hand-eye calibrations. Many experimental results of visual servoing with the proposed method has been done, and they show the feasibility and applicability of our newly proposed to a simultaneous control of camera self-motion and object motions.

  25. Study on Image Understanding using Shadow and Shading Competitive

    1994/04 - 1999/03

  26. Study of Remote Visual Servoing Control by Pseudo Stereo Method

    DEGUCHI Koichiro, OKATANI Takayuki, HONTANI Hidekata

    Offer Organization: Japan Society for the Promotion of Science

    System: Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)

    Category: Grant-in-Aid for Scientific Research (B)

    1997 - 1999

    More details Close

    In this research, we established a visual servoing method for remote control of robot employing the concept of pseudo-stereo. This method is based on stereo image principle to replay the same motions of robot currently controlled as those in a motion image transmitted from a remote place or saved in old record. Our purposes of this research were to construct its theoretical base and to realize a remote control system based on the theory including methodology for effective image transmission and robot control. We discussed with specialists in commercial factories of robots and exchanged ideas to improve the feasibility of our system. By these discussions, we acknowledged the availability of our visual servoing system in production lines and other fields of manufacturing. To summarize our three year research project ; 1. We constructed an application oriented system for visual remote control of robots based on the concept of the pseudo stereo, and verified its feasibility by many experiments simulating its application of production field. 2. From the experimental tests on performances of robot control and image processing, we verified that the system could be constructed using commercially available high-performance CPU, needless any high cost specialized hardware for the required image processing. It promised the real applications in wide production fields. 3. We summarized the theory of an application of the pseudo stereo as well as robot navigation, robot control, and remote image transmission techniques realized in the research, and published at international journals, proceedings of international conferences and technical reports.

Show all Show first 5

Other 2

  1. 認識の数理モデルと高階・多層確率場による高次元実データ解析

  2. 極限環境を対象とした頑健なロボット視覚の実現

    More details Close

    極限環境を対象とした頑健なロボット視覚を実現する.