Advance

ABE Koji

    Department of Informatics　Associate Professor
Last Updated :2024/04/25

Researcher Information

Degree

  • Ph.D. in Engineering（Kanazawa University）

URL

J-Global ID

Research Interests

  • 感性情報処理   マルチメディアデータベース   パターン認識   情報検索   Medical Image Processing   Multimedia Database   Pattern Recognition   Visual Information Retreival   

Research Areas

  • Informatics / Human interfaces and interactions
  • Informatics / Sensitivity (kansei) informatics
  • Informatics / Database science
  • Environmental science/Agricultural science / Agricultural environmental and information engineering
  • Life sciences / Medical systems
  • Informatics / Perceptual information processing

Academic & Professional Experience

  • 2022/04 - Today  Cyber Informatics Research Institute, Kindai University
  • 2022/04 - Today  Kindai UniversityFaculty of Informatics Department of InformaticsAssociate Professor
  • 2009/04 - Today  Graduate School of Kindai UniversityGraduate School of Science and Engineering准教授
  • 2010/04 - 2022/03  Kindai UniversityFaculty of Science and Engineering, Department of Informatics准教授
  • 2006 - 2010  Kindai UniversityFaculty of Science and Engineering, Department of Informatics
  • 2003 - 2006  金沢工業大学情報系情報工学科 講師
  • 2002 - 2003  Northenbria UniversityInstitute for Imge Data ResearchHonorary Research Fellow
  • 2001 - 2002  Kanazawa UniversityFaculty of Engineering
  • 1999 - 2001  金沢経済大学総合情報センター 助手

Education

  • 1998/04 - 2001/03  Grad. Sch. of Kanazawa University  自然科学研究科  数理情報科学専攻
  • 1996/04 - 1998/03  Grad. Sch. of Kogakuin University  工学研究科  電気工学専攻
  • 1992/04 - 1996/03  Kogakuin University  工学部  Department of Computer Science and Communication Engineering

Association Memberships

  • 電気学会   画像電子学会   IEEE   電子情報通信学会   

Published Papers

  • Ikuto Ogawa; Masayuki Otani; Koji Abe; Nobukazu Iguchi; Hitoshi Habe
    2022 IEEE 8th World Forum on Internet of Things (WF-IoT) IEEE 1 - 2 2022/10 [Refereed]
  • M. Yamamuro; Y. Asai; N. Hashimoto; N. Yasuda; H. Kimura; T. Yamada; M. Nemoto; Y. Kimura; H. Handa; H. Yoshida; K. Abe; M. Tada; H. Habe; T. Nagaoka; S. Nin; K. Ishii; Y. Lee
    Proc. SPIE 12286, 16th International Workshop on Breast Imaging SPIE (122860T) 2022/07 [Refereed]
  • Mika Yamamuro; Yoshiyuki Asai; Naomi Hashimoto; Nao Yasuda; Hiorto Kimura; Takahiro Yamada; Mitsutaka Nemoto; Yuichi Kimura; Hisashi Handa; Hisashi Yoshida; Koji Abe; Masahiro Tada; Hitoshi Habe; Takashi Nagaoka; Seiun Nin; Kazunari Ishii; Yohan Kondo
    Biomedical Physics & Engineering Express IOP Publishing 8 (4) 045016 - 045016 2022/06 [Refereed]
     
    Abstract This study investigates the equivalence or compatibility between U-Net and visual segmentations of fibroglandular tissue regions by mammography experts for calculating the breast density and mean glandular dose (MGD). A total of 703 mediolateral oblique-view mammograms were used for segmentation. Two region types were set as the ground truth (determined visually): (1) one type included only the region where fibroglandular tissue was identifiable (called the ‘dense region’); (2) the other type included the region where the fibroglandular tissue may have existed in the past, provided that apparent adipose-only parts, such as the retromammary space, are excluded (the ‘diffuse region’). U-Net was trained to segment the fibroglandular tissue region with an adaptive moment estimation optimiser, five-fold cross-validated with 400 training and 100 validation mammograms, and tested with 203 mammograms. The breast density and MGD were calculated using the van Engeland and Dance formulas, respectively, and compared between U-Net and the ground truth with the Dice similarity coefficient and Bland–Altman analysis. Dice similarity coefficients between U-Net and the ground truth were 0.895 and 0.939 for the dense and diffuse regions, respectively. In the Bland–Altman analysis, no proportional or fixed errors were discovered in either the dense or diffuse region for breast density, whereas a slight proportional error was discovered in both regions for the MGD (the slopes of the regression lines were −0.0299 and −0.0443 for the dense and diffuse regions, respectively). Consequently, the U-Net and ground truth were deemed equivalent (interchangeable) for breast density and compatible (interchangeable following four simple arithmetic operations) for MGD. U-Net-based segmentation of the fibroglandular tissue region was satisfactory for both regions, providing reliable segmentation for breast density and MGD calculations. U-Net will be useful in developing a reliable individualised screening-mammography programme, instead of relying on the visual judgement of mammography experts.
  • Takashi Nagaoka; Takenori Kozuka; Takahiro Yamada; Hitoshi Habe; Mitsutaka Nemoto; Masahiro Tada; Koji Abe; Hisashi Handa; Hisashi Yoshida; Kazunari Ishii; Yuichi Kimura
    Advanced Biomedical Engineering Japanese Society for Medical and Biological Engineering 11 76 - 86 2022/03 [Refereed]
  • A Feature Value for Measuring Progression of Gastric Atrophy Utilizing the Distribution of Folds in Gastric X-ray Images
    K. Abe; K. Ito; M. Minami
    Proc. of the 7th IIEEJ International Conference on Image Electronics and Visual Computing (A6-6) 2021/09 [Refereed]
  • K. Abe; T. Maetani; M. Minami
    Proc. of the 7th IIEEJ International Conference on Image Electronics and Visual Computing 2021 (A1-1) 2021/09 [Refereed]
  • S. Abe; T. Takagi; S. Torisawa; K. Abe; H. Habe; N. Iguchi; K. Takehara; S. Masuma; H. Yagi; T. Yamaguchi; S. Asaumi
    Aquacultural Engineering Elsevier BV 93 102146 - 102146 0144-8609 2021/05 [Refereed]
  • M. Yamamuro; Y. Asai; N. Hashimoto; N. Yasuda; T. Yamada; M. Nemoto; Y. Kimura; H. Handa; H. Yoshida; K. Abe; M. Tada; H. Habe; T. Nagaoka; Y. Ozaki; S. Nin; K. Ishii; Y. Lee
    Proc. of SPIE 11597, Medical Imaging 2021: Computer-Aided Diagnosis SPIE-INT SOC OPTICAL ENGINEERING 11597 (115972V) 0277-786X 2021/02 [Refereed]
     
    In individualized screening mammography, a breast density is important to predict potential risks of breast cancer incidence and missing lesions in mammographic diagnosis. Segmentation of the mammary gland region is required when focusing on missing lesions. A deep-learning method was recently developed to segment the mammary gland region. A large amount of ground truth (prepared by mammary experts) is required for highly accurate deep-learning practice; however, this work is time- and labor-intensive. To streamline the ground truth in deep learning, we investigated a difference in acquired mammary gland regions among multiple radiological technologists having various experience and reading levels, who shared the criteria on segmentation. If we can ignore a skill level for image reading, we can increase a number of training images. Three certified radiological technologists segmented the mammary gland region in 195 mammograms. The degree of coincidence among them was assessed with respect to seven factors which indicated the feature of segmented regions including the breast density and mean glandular dose, using Student's t-test and Bland-Altman analysis. The assessments made by the three radiological technologists were consistent considering all factors, except the mean pixel value. Thus, we concluded that the ground truths prepared by multiple practitioners with different experiences can be accepted for the segmentation of the mammary gland region and they are applicable for training images if they stringently share the criteria on the segmentation.
  • Koji Abe; Shinichiro Kuroda; Hitoshi Habe
    Sensors and Materials MYU K.K. 32 (11) 3581 - 3581 0914-4935 2020/11 [Refereed]
  • Koji Abe; Koki Miyake; Hitoshi Habe
    Sensors and Materials MYU K.K. 32 (11) 3595 - 3595 0914-4935 2020/11 [Refereed]
  • A Scoring Model of Figural Goodness and Its Application to Contour Completion Problems
    T. Hayashi; K. Abe
    Proc. of IEEE International Symposium on Multimedia 303 - 306 2019/12 [Refereed]
  • Extraction of the Essential Region in Computer-aided Diagnosis of Gastric Atorophy Using Gastric X-ray Images
    K. Abe; K. Shirakawa; M. Minami; K. Yoshikawa
    Information Engineering Express 5 (2) 99 - 108 2019/11 [Refereed][Invited]
  • Improvement of the System for Measuring VDT Working Hours Using a Webcam
    K. Abe; A. Soda; M. Minami
    Proc. of the 4th International Conference on Frontiers of Signal Processing 130 - 134 2018/09 [Refereed]
  • Features for Evaluating Gastric Atrophy Using X-ray Images
    K. Abe; K. Shirakawa; M. Minami; D. Miura
    Proc. of the 4th International Conference on Frontiers of Signal Processing 94 - 99 2018/09 [Refereed]
  • Koji Abe; Masaru Tanaka; Hitoshi Habe; Yoshiaki Taniguchi; Nobukazu Iguchi
    Proc. of the 4th International Conference on Frontiers of Signal Processing 105 - 109 2018/09 [Refereed]
  • Automated Extraction of the Essential Region in Computer-aided Diagnosis of Helicobacter Pylori Infection Using Gastric X-ray Images
    K. Abe; K. Shirakawa; M. Minami; K. Yoshikawa
    Proc. of International Congress on Advanced Applied Informatics 626 - 629 2018/07 [Refereed]
  • Similarity Retrieval of Trademark Images by Vector Graphics Based on Shape Characteristics of Components
    K. Abe; H. Morita; T. Hayashi
    Proc. of 2018 10th International Conference on Computer and Automation Engineering 82 - 86 2018/02 [Refereed]
  • Similarity Retrieval of 3D Models with Query by Clay Sketch
    T. Hayashi; M. Hirose; K. Suzuki; K. Abe
    Proc. of 2016 IEEE International Symposium on Multimedia 173 - 176 2016/12 [Refereed]
  • A Similarity Retrieval of Trademark Images Considering Similarity for Local Objects Using Vector Images
    H. Morita; K. Abe; T. Hayashi
    Proc. of the 8th International Conference on Signal Processing Systems 79 - 83 2016/11 [Refereed]
  • Features for Discriminating Helicobacter Pylori Infection from Gastric X-ray Images
    K. Abe; D. Miura; M. Minami
    Proc. of the 8th International Conference on Signal Processing Systems 31 - 35 2016/11 [Refereed]
  • Koji Abe; Ryota Shimizu; Hitoshi Habe; Yoshiaki Taniguchi; Nobukazu Iguchi
    Proc. of the 8th International Conference on Signal Processing Systems 6 - 10 2016/11 [Refereed]
  • Stability Improvement of Indirect Matching for Music Information Retrieval
    T. Hayashi; N. Ishii; M. Ishimori; K. Abe
    Proc. of IEEE International Symposium on Multimedia 229 - 232 2015/12 [Refereed]
  • Improvement of Image and Video Matting with Multiple Reliability Maps
    T. Hayashi; M. Ishimori; N. Ishii; K. Abe
    Proc. of IEEE International Symposium on Multimedia 225 - 228 2015/12 [Refereed]
  • Abe Koji; Takagi Hidenori; Minami Masahide; Tian Haiyan
    The Journal of the Institute of Image Electronics Engineers of Japan The Institute of Image Electronics Engineers of Japan 44 (3) 493 - 501 2186-716X 2015/07 [Refereed]
     
    This paper presents features for measuring the congestive extent as a means for diagnosing internal hemorrhoids using endoscopic images. A computer-aided diagnosis for the hemorrhoids could be effective for young or even general practitioners as a second opinion because diagnosis of hemorrhoids needs much experience. Considering characteristics of the hemorrhoids, the proposed method measures degree of the congestion in the images and the size of the congestive regions as abnormalities of the hemorrhoids. Experimental results of the discrimination between normal and abnormal cases using the proposed abnormalities for 204 images (96 normal cases and 108 abnormal cases) have shown that the abnormalities are well effective to measure degree of the congestion in internal hemorrhoids.
  • Sound Classification for Hearing Aids Using Time-frequency Images
    K. Abe; H. Masaki; H. Tian
    Journal of Basic and Applied Physics 3 (4) 159 - 166 2014/11 [Refereed]
  • Koji Abe; Minori Yama; Masahide Minami; Haiyan Tian
    Proceedings of the IASTED International Conference on Modelling, Simulation and Identification, MSI 2014 Acta Press 121 - 127 2014 [Refereed]
     
    This paper presents a system for giving the warning information to the PC user in companies before going into working in front of the PC for long hours. This system is utilized in companies to avoid long VDT work. Although it would be not difficult to monitor the PC user by using motionsensing devices such as Kinect whether the user is doing VDT work, this way is not practical due to the high cost for companies. Assuming large-volume introduction of the system into companies, the proposed system monitors the PC user using a common webcam considering cost performance. The proposed method extracts just one feature for measuring area of the user's face and head from images captured at the interval of 1 min. and then discriminates whether the user appears in the camera frame or not. Besides, the proposed method measures hours of VDT work by counting the result of the discrimination for the user's appearance at every minute and gives a warning information to the user before the user is doing VDT work for long hours. Experimental results that accuracy ratios for the discrimination have been 93.9% on average have shown thatperformance of the proposed method is high enough.
  • Koji Abe; Hiroyoshi Masaki; Haiyan Tian
    Proceedings of the IASTED International Conference on Modelling, Simulation and Identification, MSI 2014 Acta Press 115 - 120 2014 [Refereed]
     
    This paper presents features of sound data for a sound classification equipped for hearing aids. The features are extracted by using image processing techniques to timefrequency images. As an application of hearing aids in mind, four classes of "classical music", "speech", "multitalker noise" and "speech in the noise" are prepared in order to classify the input signal of a hearing aid into useful classes. Although there are several possible ways to figure out which class the current input signal belongs to, an approach from image processing is utilized to find out appropriate features because 2D image (time-frequency image) can contain multifaceted information compared to 1D information (waveform or frequency response of sound), and can be regarded as comprehensive data. It is found that eight features are required to meet a certain quality of sound classification according to our investigation. Experimental results of the sound classification by some clustering machines using the proposed features have shown that accuracy of the classification was more than 95 % with every clustering machine.
  • Features for Measuring the Congestive Extent of Internal Hemorrhoids in Endoscopic Images
    K. Abe; H. Takagi; M. Minami; H. Tian
    Proc. of the 11th Australasian Data Mining Conference 2013/11 [Refereed]
  • Extraction of Essential Region in Gastric Area for Diagnosing Gastric Cancer Using Double Contrast X-ray Images
    K. Abe; H. Nakagawa; M. Minami; H. Tian
    Proc. of the 11th Australasian Data Mining Conference 2013/11 [Refereed]
  • HAYASHI Takahiro; UEKI Syunsuke; ABE Koji
    The IEICE transactions on information and systems (Japanese edition) The Institute of Electronics, Information and Communication Engineers J96-D (11) 2804 - 2814 1880-4535 2013/11 [Refereed]
     
    ラスタベクタ変換は画像編集,画像認識,画像検索等様々な分野において広く利用されている.一方で,既存のラスタベクタ変換手法は,人間の視知覚特性を考慮していないため,複数のオブジェクトが重なっている場合に,オブジェクトを正しく分割できないという限界があった.この問題を解決するために,本論文では,部分的に遮蔽された図形の輪郭線を補完する手法を提案する.提案手法では,部分的に遮蔽された図形が線対称図形であると判断される場合に,候補対称軸を設定し,その軸で図形を折り返すことで,欠損部分の輪郭線補完を試みる.折り返した輪郭線と元の輪郭線とで重なる部分が大きくなる場合に,候補対称軸を線対称軸として検出し,このときの折り返しにより得られる結果を補完結果として出力する.提案手法の有効性を評価するために,部分的に遮蔽された図形に対して,提案手法による補完が正解(人間による補完の結果)を出力できるかどうか調査した.また,既存手法である曲線による補完,直線による補完の結果とも比較した.その結果,既存手法と提案手法を適切に組み合わせ,線対称図形であると判断できる図形に対しては提案手法を適用し,それ以外の図形に対してのみ既存手法を適用することで,人間による補完結果により近い補完結果を出力できることを示した.
  • A Method for Extracting Grouping Areas of Good Continuity Parts in Trademark Images
    B. Befane; K. Abe; T. Hayashi
    Proc. of the 27th Image and Vision Computing New Zealand 295 - 300 2012/11 [Refereed]
  • KAJIWARA Yusuke; MURATA Hiroaki; KIMURA Haruhiko; ABE Koji
    IEEJ Transactions on Electronics, Information and Systems The Institute of Electrical Engineers of Japan 132 (4) 555 - 560 0385-4221 2012/04 [Refereed]
     
    As a communication support tool for cases of amyotrophic lateral sclerosis (ALS), researches on eye gaze human-computer interfaces have been active. However, since voluntary and involuntary eye movements cannot be distinguished in the interfaces, their performance is still not sufficient for practical use. This paper presents a high performance human-computer interface system which unites high quality recognitions of horizontal directional eye movements and voluntary blinks. The experimental results have shown that the number of incorrect inputs is decreased by 35.1% in an existing system which equips recognitions of horizontal and vertical directional eye movements in addition to voluntary blinks and character inputs are speeded up by 17.4% from the existing system.
  • 中村宗広; 梶原祐輔; 金谷二郎; 阿部孝司; 木村春彦
    画像電子学会誌 40 (6) 1009 - 1016 1348-0316 2011/11 [Refereed]
  • KANAYA Jiro; NAKAMURA Munehiro; ARAKI Masahiro; YOKAWA Hiroshi; ABE Koji; KIMURA Haruhiko
    IEEJ Transactions on Electronics, Information and Systems The Institute of Electrical Engineers of Japan 131 (4) 726 - 735 0385-4221 2011/04 [Refereed]
     
    With increase of aged people in Japan, cases of diseases in bladder are increased. Hence, incidents of death due to bladder cancer are increasing nowadays. Cystoscope is the most essential tool in examination of bladder. Although CT, MRI, and ultrasonography are painless examinations, they lack in accuracy. There are two kinds of cystoscopes namely rigid and flexible. The examination with the latter one is painless and is used widely. In handling of flexible cystoscope, medical doctors need to observe the whole inner wall of bladder regulating the direction of the tip, rotation on the cystoscope's axis, and depth of insertion. However, beginners handling this equipment tend to lose track of the observation resulting in poor characteristics of images obtained from the cystoscope. Sometimes they would pass over some parts in the whole bladder. As the preprocessing of a system for thorough observation, this paper proposes a method for estimating the handling techniques of the cystoscope using neural networks to images obtained from the cystoscope.
  • Koji Abe; Hidetoshi Sakaue; Takatoshi Okuno; Kiyoaki Terada
    2011 IEEE PACIFIC RIM CONFERENCE ON COMMUNICATIONS, COMPUTERS AND SIGNAL PROCESSING (PACRIM) IEEE 719 - 724 2011 [Refereed]
     
    This paper presents a sound classification method for hearing aids based on time-frequency image processing. As an application of hearing aids in mind, four classes of "Speech", "Speech in noise", "Noise" and "Classical music" are prepared in order to classify the input signal of a hearing aid into useful classes. Although there are several possible ways to figure out which class the current input signal belongs to, an approach from image processing is utilised to find out several appropriate features because 2D image (time-frequency image) can contain multifaceted information compared to 1D information (waveform or frequency response of sound), and can be regarded as comprehensive data. It is found that eleven features are required to meet a certain quality of sound classification according to our investigation.
  • Koji Abe; Tetsuya Nobuoka; Masahide Minami
    2011 IEEE PACIFIC RIM CONFERENCE ON COMMUNICATIONS, COMPUTERS AND SIGNAL PROCESSING (PACRIM) IEEE 708 - 713 2011 [Refereed]
     
    In a mass screening for gastric cancer, diagnosticians read several hundred stomach X-ray pictures at a time. The existing systems of computer-aided diagnosis for the cancer mark the location of lesions appeared in X-ray images. However, the systems do not reduce the hard labor due to lack of the accuracy in the marking. Besides, to diagnose characteristics of legions, diagnosticians have to directly read X-ray pictures of abnormal cases even if the systems could show the location precisely. For the sake of decreasing the number of reading the pictures in the mass screenings, the proposed method discriminates normal cases using stomach X-ray images. In normal cases, folds on the stomach wall appear in parallel. Therefore, the proposed method measures features of parallelism extracting the folds from X-ray images. Experimental results of the discriminations for 43 images including 11 abnormal cases have shown that the proposed features are well effective for recognizing normal cases.
  • ABE Koji; MINAMI Masahide; NAKAMURA Munehiro
    The IEICE transactions on information and systems The Institute of Electronics, Information and Communication Engineers J94-D (1) 395 - 408 1880-4535 2011/01 [Refereed]
     
    じん肺の診断では胸部X線写真の読影が難しく,これまでいくつかコンピュータによりじん肺を定量的に診断する方法が提案されてきた.しかし,既存方法では胸部X線写真を鮮明にディジタル化するスキャナが高価でモバイル性に欠けていることから,じん肺に対するコンピュータ診断はいまだ実用化に至っていない.そこで,本論文では,市販CCDスキャナで取り込んだ胸部X線像に対するじん肺コンピュータ診断に有効なじん肺異常度を提案する.市販CCDスキャナで取り込んだ胸部X線像は,フィルム専用スキャナで取り込んだものと比べ画質が著しく劣化する.本手法では,市販CCDスキャナで取り込んだ胸部X線像に対して,ユーザがタブレットペンで肋骨エッジを描画して肋骨領域と肋間領域を決定し,それらの領域からじん肺の症状の度合を特徴量として抽出する.本手法に対する評価実験では,厚生労働省が提供しているじん肺標準写真9枚と健常者の胸部X線写真47枚に対し,本手法により抽出された特徴量を用いて種々の分類器で健常と異常の判別を行った.実験の結果,健常・異常判別の再現率は83.3%以上,適合率は98.0%以上となり,本手法の有効性を確認した.
  • Brain-computer Interface for Assisting Decision-making on Individual Preference by Switching Support Vector Machines
    T. Misawa; S. Takano; K. Abe; T. Shimokawa; S. Hirobayashi
    International Journal of Computational Science 4 (6) 477 - 490 2010/12 [Refereed]
  • Hiromasa Iguchi; Koji Abe; Tadanobu Misawa; Haruhiko Kimura; Yoshimasa Daido
    ELECTRONICS AND COMMUNICATIONS IN JAPAN SCRIPTA TECHNICA-JOHN WILEY & SONS 92 (10) 49 - 60 1942-9533 2009/10 [Refereed][Invited]
     
    Based oil Gestalt principles, in this paper we model Features For measuring the degree of similarity between combinations of image components, and grouping areas in trademark images are recognized. This investigation can be used for content-based image retrieval from the perspective of mirroring human perception of images. The Features of proximity, shape similarity, closure, and good continuation are extracted from every combination of two components in an image. After that, based on the decision results, a grouping pattern for the query is fixed. Besides changing combinations of features, the proposed method can output multiple grouping patterns. In the experiments, we evaluated the proposed method on 74 test images, comparing outputs produced by the proposed method and grouping for the test images obtained from questionnaires filled Out by 104 participants. (C) 2009 Wiley Periodicals, Inc. Electron Comm Jpn, 92(10): 49-60, 2009; Published in Wiley InterScience (www.interscience. wiley.com). DOI 10.1002/ccj.10007
  • 癒し型ペットロボットの飼い主判別手法の改善
    南保英孝; 木村春彦; 原未來; 阿部孝司; 田嶋拓也
    電気学会論文誌C 129-C (9) 1662 - 1668 2009/09 [Refereed]
  • Quantitative Evaluation of Pneumoconiosis in Chest Radiographs Obtained with a CCD Scanner
    M. Nakamura; K. Abe; M. Minami
    Proc. of the 2nd International Conference on the Applications of Digital Information and Web Technologies 673 - 678 2009/08 [Refereed]
  • Hidetaka Nambo; Haruhiko Kimura; Mirai Hara; Koji Abe; Takuya Tajima
    IEEJ Transactions on Electronics, Information and Systems Institute of Electrical Engineers of Japan 129 (9) 5 - 1668 1348-8155 2009 [Refereed]
     
    In order to decrease human stress, Animal Assisted Therapy which applies pets to heal humans is attracted. However, since animals are insanitary and unsafe, it is difficult to practically apply animal pets in hospitals. For the reason, on behalf of animal pets, pet robots have been attracted. Since pet robots would have no problems in sanitation and safety, they are able to be applied as a substitute for animal pets in the therapy. In our previous study where pet robots distinguish their owners like an animal pet, we used a puppet type pet robot which has pressure type touch sensors. However, the accuracy of our method was not sufficient to practical use. In this paper, we propose a method to improve the accuracy of the distinction. The proposed method can be applied for capacitive touch sensors such as installed in AIBO in addition to pressure type touch sensors. Besides, this paper shows performance of the proposed method from experimental results and confirms the proposed method has improved performance of the distinction in the conventional method. © 2009 The Institute of Electrical Engineers of Japan.
  • Takahiro Hayashi; Koji Abe; Debabrata Roy; Rikio Onai
    International Journal of Business Intelligence and Data Mining 4 (1) 62 - 77 1743-8187 2009 [Refereed]
     
    This paper presents a method for discriminating between personal and non-personal web pages. The method can support surveys of personal opinions about products and services. In the proposed method, subjective expressions are extracted from pages and then the pages are scored by quantitatively evaluating the subjectivity in the pages. We have evaluated performances of the proposed method using 1200 web pages collected from four categories of product, tourist spot, restaurant, and movie. Comparing the performances of the proposed method with categorisations by a general search engine, we have confirmed that the performances have been significantly better in every category. Copyright © 2009, Inderscience Publishers.
  • Retrieval of 2D Vector Images by Matching Weighted Feature Points
    T. Hayashi; K. Abe; R. Onai
    Proc. of the 2008 IEEE International Conference on Image Processing 961 - 964 2008/10 [Refereed]
  • TAJIMA Takuya; KIMURA Haruhiko; ABE Takehiko; ABE Koji; NAKAMOTO Yoshinori
    IEEJ Transactions on Electronics, Information and Systems The Institute of Electrical Engineers of Japan 128 (9) 1470 - 1479 0385-4221 2008/09 [Refereed]
     
    Information about customers in shops plays an important role in marketing analysis. Currently, in convenience stores and supermarkets, the identification of customer's gender is examined by clerks. On the other hand, gender identification systems using camera images are investigated. However, these systems have a problem of invading human privacies in identifying attributes of customers. The proposed system identifies gender by using infrared area scanners and Bayesian network. In the proposed system, since infrared area scanners do not take customers' images directly, invasion of privacies are not occurred. The proposed method uses three parameters of height, walking speed and pace for humans. In general, it is shown that these parameters have factors of sexual distinction in humans, and Bayesian network is designed with these three parameters. The proposed method resolves the existent problems of restricting the locations where the systems are set and invading human privacies. Experimental results using data obtained from 450 people show that the identification rate for the proposed method was 91.3% on the average of both of male and female identifications.
  • Performance Comparison of Flux-Concentration Type and Conventional Type Tubular Linear Induction Motor Using Three-Phase Equivalent Circuit Parameters
    D. Roy; K. Abe; B. Basak
    Journal of the Institution of Engineers (Electrical Engineering Division) 89 27 - 33 2008/06 [Refereed]
  • Retrieval of Personal Web Documents by Extracting Subjective Expressions
    T. Hayashi; K. Abe; R. Onai
    Proc. of the 2008 IEEE International Symposium on Mining the Asian Web 1187 - 1192 2008/03 [Refereed]
  • Investigation of the Performances of a Passive Magnetic Fault Current Limiter
    A. B. Choudhury; D. Roy; K. Abe
    Journal of the Institution of Engineers (Electrical Eng. Division) 88 32 - 35 2008/03 [Refereed]
  • 自己代理強化学習による交通信号制御
    参沢匡将; 阿部孝司; 下川哲矢; 木村春彦
    情報処理学会論文誌 49 (1) 445 - 456 2008/01 [Refereed]
  • Vector Image Segmentation for Content-Based Vector Image Retrieval
    T. Hayashi; R. Onai; K. Abe
    Proc. of the 7th IEEE International Conference on Computer and Information Technology 695 - 700 2007/10 [Refereed]
  • 田嶋拓也; 阿部武彦; 阿部孝司; 南保英孝; 木村春彦
    日本生産管理学会論文誌 日本生産管理学会 14 (1) 133 - 138 1341-528X 2007/10 [Refereed]
     
    近年の情報流通の発達に伴い, 価値観が多様化しているといった現状があり, CRMなどが重要化している中, マーケティングや経営問題解決における顧客情報がより一層重要化している。コンビニエンスストアやスーパーマーケットなどの店舗ではさまざまな顧客情報を取得し, マーケティングに活用しているが, 顧客グループ数や顧客グループの属性や特性を取得することは容易でないといった問題点もある。そこで本研究では, 顔画像などの個人情報を取得しないことを前提とした顧客グループ認知システムを提案し, 開発を行うことを目的とする。
  • ABE Koji; NENDAI Mitsuhiro; HAYASHI Takahiro; KIMURA Haruhiko
    The Journal of the Institute of Image Electronics Engineers of Japan The Institute of Image Electronics Engineers of Japan 36 (5) 701 - 709 0285-9831 2007/09 [Refereed]
     
    Contents of sports tickers in broadcasted sports videos could be efficient as queries in analyzing the video contents. However, since sports tickers are displayed with various colors and brightness, it is difficult to recognize inner objects of the sports tickers only according to the brightness distribution. Besides, in order to recognize what the objects represent, it is indispensable to investigate the change of these objects according to time traveling. In this paper, the proposed method can recognize locations of the out ticker, score tickers, and runner markers in the sports ticker of broadcasted baseball videos analyzing the brightness histograms obtained from every of grayscale and color images, and the change of the sports ticker with the correlation coefficient. Experimental results with four broadcasted baseball videos have shown that the proposed method can exactly work in the case when the out ticker is represented by a letter.
  • K. Abe; K. Miyashita; H. Murakami; T. Hayashi; H. Tian
    Proc. of the 9th IASTED International Conference on Signal and Image Processing ACTA PRESS ANAHEIM 242 - 247 1482-7921 2007/08 [Refereed]
     
    Sports tickers in broadcasted sports videos could be useful as a query in analyzing the video contents. However, since sports tickers are displayed with various colors and brightness, it is difficult to recognize inner objects of the sports tickers according to only the brightness distribution. Besides, in order to recognize what the objects represent, it is indispensable to investigate change of these objects according to time progress. In this paper, the proposed method can recognize locations of the out ticker, score tickers, and runner marks in the sports ticker of broadcasted baseball videos analyzing histograms obtained from grayscale and RGB color values in frames, and change of the sports ticker with the correlation coefficient. Experimental results with four broadcasted baseball videos have shown that the proposed method can exactly work in the case when the out ticker is represented by a number.
  • ゲシュタルト心理学に基づく抽象図形の群化領域認識
    井口弘将; 阿部孝司; 参沢匡将; 木村春彦; 大洞喜正
    電気学会論文誌Ｃ, vol.127, no.6, pp.844-853, Jun. 2007. 127 (6) 844 - 853 2007/06 [Refereed]
  • K. Abe; T. Misawa; M. Nendai; H. Tian
    IEEJ Trans. on Electrical and Electronic Engineering 1 (4) 374 - 382 1931-4973 2006/11 [Refereed]
  • Figure-Ground Separation for Vector Graphics with Contour Detection
    T. Hayashi; K. Shibata; R. Onai; K. Abe
    Proc. of the 10th IASTED International Conference on Internet and Multimedia Systems and Applications 192 - 197 2006/08 [Refereed]
  • K Abe; H Iguchi; HY Tian; D Roy
    IEICE TRANSACTIONS ON INFORMATION AND SYSTEMS IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG E89D (6) 1798 - 1805 0916-8532 2006/06 [Refereed]
     
    According to the Gestalt principals, this paper presents a recognition method of grouping areas in trademark images modeling features for measuring the attraction degree between couples of image components. This investigation would be used for content-based image retrieval from the view of mirroring human perception for images. Depending on variability in human perception for trademark images, the proposed method finds grouping areas by calculating Mahalanobis distance with the features to every combination of two components in images. The features are extracted from every combination of two components in images, and the features represent proximity, shape similarity, and closure between two components. In addition, changing combination of the features, plural grouping patterns are output. Besides, this paper shows the efficiency and limits of the proposed method from experimental results. In the experiments, 104 participants have perceived grouping patterns to 74 trademark images and the human perceptions have been compared with outputs by the proposed method for the 74 images.
  • 感覚概念を導入したニオイセンサによる環境認識に関する一考察
    参沢匡将; 竹井義法; 南戸秀仁; 阿部孝司; 木村春彦
    電気学会論文誌Ｅ 126 (3) 107 - 113 2006/03 [Refereed]
  • A Speedup Algorithm for Repetition of Hypothetical Reasoning
    H. Kimura; T. Misawa; K. Abe; Y. Ogoshi
    Journal of Advanced Computational Intelligence and Intelligent Informatics 10 (2) 207 - 218 2006/03 [Refereed]
  • TACHI Yoshinobu; TSUJISAWA Narihisa; MISAWA Tadanobu; ABE Koji; KIMURA Haruhiko; NAMBO Hidetaka
    The IEICE transactions on information and systems The Institute of Electronics, Information and Communication Engineers J89-D (3) 447 - 456 1880-4535 2006/03 [Refereed]
     
    発話者の感情を認識することは,臨場感ある対話システムの開発やケアシステムの作成に役立つ.感情認識の研究には,単語や会話の意味解釈から判別するものや,発話者の音声から判別するものがある.後者は,発話者が特定話者の場合には認識率を向上させることができるが,不特定話者の場合には実用に耐え得ることができなかった.本論文では,ペットロボットをなでながら音声入力をしてもらうことにより,不特定話者の感情認識を行う方法を提案する.
  • M. Hirashita; O. Shido; K. Abe; Y. Kajiwara; S. Yokokura
    Proc. of the 3rd International Conference on Human-Environment System 2005 (Web) 417 - 421 2434-8007 2005/09 [Refereed]
  • 吉田健吾; 蟹屋敷祐介; 横内繁幸; 阿部孝司; 南保英孝; 木村春彦
    日本設備管理学会誌 日本設備管理学会 17 (3) 151 - 159 0915-5023 2005 [Refereed]
  • Koji Abe; Yoshimasa Daido; Hiromasa Iguchi; Haiyan Tian; Debabrata Roy
    Proceedings of the IASTED International Conference on Human-Computer Interaction ACTA PRESS ANAHEIM 167 - 172 2005 [Refereed]
     
    According to the Gestalt principals, in this paper, we model features for measuring the attraction degree between pairs of image components, and grouping areas in trademark images are recognized. This investigation would be used for content-based image retrieval from the view of mirroring human perception for images. Depending on variability in human perception for trademark images, the proposed method finds plural grouping areas by calculating Mahalanobis distance with the features. The features are extracted from every combination of two components in an image. We have evaluated the proposed method on 104 test images in experiments of comparing between outputs by the proposed method and by human perception, and we have got the concordance rate of 81.99%.
  • K Abe; D Roy; JP Eakins
    KNOWLEDGE-BASED INTELLIGENT INFORMATION AND ENGINEERING SYSTEMS, PT 1, PROCEEDINGS SPRINGER-VERLAG BERLIN 3213 614 - 619 0302-9743 2004 [Refereed]
     
    According to the Gestalt principals, we model features for representing the attraction degree between pairs of image regions, and, grouping areas in trademarks are recognized. This investigation would be used for content-based image retrieval systems from the view of mirroring human perception for trademark images. Depending on variability in human perception for trademark images, the proposed method finds grouping areas considering combinations of the features. We have evaluated the effectiveness of the proposed method on 36 test images of abstract trademarks from the Japan Trademark Registry. Experimental results have shown over 80% in the images have been the same as humans perception.
  • A System for Drawing Monochrome Portraits for Criminal Detection
    K. Abe; T. Okamine; S. Tanaka; T. Abe; H. Kimura
    Proc. of The 4th Asia-Pacific Conference on Industrial Engineering and Management Systems 1181 - 1185 2002/12 [Refereed]
  • 白黒反転商標の類似検索における前処理
    阿部孝司; 木村春彦; 長嶋秀世
    電子情報通信学会論文誌Ｄ-II J83-D-II (7) 1628 - 1637 2000/07 [Refereed]

MISC

Awards & Honors

  • 2022/11 計測自動制御学会システム・情報部門学術講演会 研究奨励賞
     内視鏡画像を用いた内痔コンピュータ診断に有効なチューブ領域検出 
    受賞者: 安福翔;阿部孝司;南昌秀
  • 2021/04 令和2年電気関係学会関西連合大会 奨励賞
     内痔診断のための内視鏡画像内のチューブ領域除去手法の改善 
    受賞者: 三宅航暉;阿部孝司;南昌秀
  • 2021/04 映像情報メディア学会関西支部 令和2年電気関係学会関西連合大会 優秀論文発表賞
     じん肺X線像における粒状影抽出改善を導入したじん肺診断支援システム 
    受賞者: 黒田真一朗;阿部孝司;南昌秀
  • 2020/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     動画を用いた生簀内クロマグロ稚魚の尾数推定 
    受賞者: 覚野陽生;阿部孝司;波部斉;谷口義明;井口信和
  • 2020/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     映像を用いた生簀内クロマグロの尾数計測 
    受賞者: 金田彩花;阿部孝司;波部斉;谷口義明;井口信和;阿部悟;鳥澤眞介;高木力
  • 2017/02 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     胸部X線CR画像を用いたじん肺自動診断支援システムの開発 
    受賞者: 多田圭汰;阿部孝司;南昌秀;田海燕
  • 2016/06 The Institute of Image Electronics Engineers of Japan Best Paper Award
     内視鏡画像を用いた内痔コンピュータ画像診断に有効なうっ血特徴量 
    受賞者: 阿部孝司;高木秀典;南昌秀;田海燕
  • 2016/04 電気関係学会関西支部 電気関係学会関西支部大会 奨励賞
     胃X線二重造影像を用いたHelicobacter pylori 感染診断に有効な特徴量 
    受賞者: 三浦大輝;阿部孝司;南昌秀;田海燕
  • 2016/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     商標図形に存在する連続性要因を持つ群化領域の認識 
    受賞者: 宇佐見光平;阿部孝司;林貴宏;田海燕
  • 2015/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     魚の遊泳映像における瞬発的遊泳時の検出 
    受賞者: 佐伯亮;阿部孝司;波部斉;井口信和
  • 2014/02 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     Webカメラを用いたVDT作業時間の自動測定システム 
    受賞者: 山稔;阿部孝司;南昌秀;田海燕
  • 2014/02 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     ベクトル集中度フィルタによる胃ひだ抽出と健常胃判別 
    受賞者: 福田綾乃;阿部孝司;南昌秀;田海燕
  • 2014/02 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     補聴器への搭載を想定した画像処理手法による音源方向推定 
    受賞者: 正木博善;阿部孝司;田海燕
  • 2013/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     モーションキャプチャによる空手演武の評価に関する基礎検討 
    受賞者: 中井英貴;阿部孝司;南昌秀
  • 2012/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     よい連続性要因を現す群化パターンの認識 
    受賞者: 貴志直征;阿部孝司
  • 2011/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     胃X線2重造影像におけるバリウム溜まりと脊柱の抽出 
    受賞者: 奥村康弘;阿部孝司;南昌秀
  • 2009/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     抽象図形内の任意２領域の平行性を測定するための特徴量の作成 
    受賞者: 小室寛子;阿部孝司
  • 2009 Journal of the Institution of Engineers (India) Certificate of Merit
     
    受賞者: D. Roy;K. Abe;B. Basak
  • 2008/03 電子情報通信学会関西支部 電子情報通信学会関西支部学生研究講演会 支部長賞「奨励賞」
     ラフペインティングを用いた風景画像検索 
    受賞者: 水津昌大;阿部孝司

Research Grants & Projects

  • Japan Society for the Promotion of Science：Grants-in-Aid for Scientific Research
    Date (from‐to) : 2020/04 -2023/03 
    Author : 阿部 孝司
     
    本研究は、画像内に存在する複数の構成要素に対し、複数の構成要素がまとまって一つに見えるという人間の知覚特性（群化知覚）を考慮して画像パターンを認識することで有効的に機能する画像情報検索を実現することである。これまで複数ある群化知覚の主要要因をモデル化し群化パターンを認識する手法を開発し、類似商標検索の性能を高めることを示してきた。本研究では、これまでの成果を更に発展させ、まだモデル化できていない要因を群化領域認識に導入し検索精度を向上させる。そして種々の社会システムに導入し群化領域認識を適用させることで精度を高めることを示す。 本研究の範囲は、(1)図形内の構成要素間の「連続性」「平行性」の群化強度を測定する特徴量の開発、(2)種々の画像解析へ(1)を応用するための技術開発、である。 (1)の連続性特徴量については、前年度、人間から得られた図形群化に関するアンケート結果を機械学習に導入し、提案した特徴量の有効性を確認した。平行性特徴量については図形商標については未だ研究段階であるが、(2)の研究内容の一部で線状画像における線の平行性を評価する特徴量を開発した。このように、(2)について、(1)の一部を導入し胃X線像の胃壁ひだ模様のパターン分類を行い自動診断支援の精度を向上させた。また、PCを用いたテレワーク作業を自動認識するシステムを開発し、ヒトの顔の存在有無を判別する前処理として(1)の一部を適用させた。 今年度の実績は次のとおりである。1.胃X線像を対象とした自動診断支援システムの開発において、線状パターンの特徴抽出に群化領域認識で用いた特徴量を一部適用し、診断支援システムの精度を向上させた。2.PCを用いたテレワークの時間測定システムを開発し、ヒトの顔の有無を判別する前処理に群化領域認識手法の一部を適用し、測定精度を向上させた。
  • 日本学術振興会：科学研究費助成事業
    Date (from‐to) : 2018/04 -2022/03 
    Author : 高木 力; 竹原 幸生; 鳥澤 眞介; 阿部 孝司; 米山 和良; 井口 信和
     
    生産物が水面下にある魚類養殖では，生産管理や成育状態を日常的に把握することが困難となっている。そのため，きめ細かい生産管理や効果的な養殖技術開発のためにICT導入による高度化が強く求められている。本研究では魚類養殖の成育・生産を効率化するシステムの構築を目指して，非接触で生簀内養殖魚の収容尾数や，成育状態を計測する基盤技術開発に取り組む。 【個体数計数システムの開発】養殖クロマグロを対象として遊泳する養殖魚魚群をワイドコンバージョンレンズを装着した小型民生品カメラを用いて20m直径の大型生簀内のほぼ全景を捉えることができた。セマンティックセグメンテーション（画像内の全画素にラベルやカテゴリを関連付ける深層学習アルゴリズム）を適用して個体画像を識別したところ，およそ5000枚の教師画像データデータセットを個体画像と背景画像の比率7:3で学習させることにより80%の検出性能を出すことができたが，システムのロバスト性を向上させるためには，個体を移動追跡するアルゴリズムを見直し，実海域での個体同士の重なりや懸濁物等のノイズ等の識別を確実化することが求められことがわかった。 【行動計測による養殖魚状態評価技術開発】個体計数システムの開発では，連続する画像間での同一個体の移動ベクトルをKC法を適用することにより推定することが可能となる。このベクトルの状態を分析することにより，養殖環境や適切な給餌量を評価することが実現できる可能性がある。そこで，連続する移動ベクトルのノルム差D，交差角θ，渦度に類似した指標ωの評価指標を算出し，養成環境の違いがこれらの指標に差をもたらすか分析した。給餌時と非給餌時では，θの値で有意な差が確認され，台風接近による風浪の激しい環境では，ノルム差やωの値に有意な差が現れた。養殖魚の飽食度合いや，波浪によるストレス度合をこれらの指標により評価できることが示唆された。
  • Japan Society for the Promotion of Science：Grants-in-Aid for Scientific Research
    Date (from‐to) : 2016/04 -2019/03 
    Author : ABE Koji
     
    To enhance performance of Content-based Image Retrieval (CBIR), the following subjects were investigated. (1) Using trademarks (black and white image), a method for recognizing grouping areas which have grouping factors of good continuity and parallelism was proposed and performance of the method was examined. Then, the method was improved by trials and errors. (2) Using vector images, a CBIR system for trademarks considering the factor of good continuity was developed. (3) In a computer-aided diagnosis for measuring stomach atrophy using X-ray images (gray image), the method of (1) was utilized to extract folds and boundary of stomach area. (4) A system for monitoring VDT work to a PC user using a webcam was developed, where the method of (1) was utilized to extract areas of human skin and eye glasses in moving images. (5) A method for detecting video scenes of burst swimming by fry was developed, where the method of (1) was utilized to analyzing fry school's motion in moving images.
  • Japan Society for the Promotion of Science：Grants-in-Aid for Scientific Research
    Date (from‐to) : 2013/04 -2016/03 
    Author : ABE Koji
     
    To enhance performance of CBIR for trademarks, the following subjects were investigated. Besides, to examine their performance more, the proposed methods were applied for pre-processing methods in medical image processing: (1) A model for measuring a grouping factor of good continuity was proposed and its performance was examined to trademarks (black and white image). Then, the model was improved by trials and errors, (2) In a computer-aided diagnosis (CAD) for stomach cancers using X-ray images (gray image), the model of (1) was utilized to extract folds and boundary of stomach area, (3) In a CAD for hemorrhoids using endoscopic images (color image), the model of (1) was utilized to extract congestive region in abnormal cases; and (4) A system for monitoring the VDT work to a PC user using a webcam was developed, where the proposed method of recognizing grouping areas including the model of (1) was utilized to extract an image feature from images obtained by the webcam.
  • Japan Society for the Promotion of Science：Grants-in-Aid for Scientific Research
    Date (from‐to) : 2010 -2011 
    Author : ABE Koji
     
    For the sake of enhancing performance of CBIRs(Content-Based Image Retrievals), the following concrete subjects were investigated. The subjects were proposed as functions of CBIRs and the proposed methods were applied to a CBIR of abstract trademark images and pre-processing methods in medical image processing in order to examine their performance as an application tool for various images. (1) A model for measuring a grouping factor of good continuity was proposed. (2) Using models for measuring proximity, shape similarity, closure, and good continuity, a method for producing grouping patterns for abstract images. (3) In a computer-aided diagnosis for pneumoconiosis using chest X-ray images, the method of(2) was applied for removing noises except abnormal parts in chest area and the method was more effective for the removal than normal methods. (4) In a computer-aided diagnosis for stomach cancers using X-ray images, the method of(1) was applied for conducting the binarization of folds in stomach area and the proposed method was effective for the binarization.
  • Japan Society for the Promotion of Science：Grants-in-Aid for Scientific Research
    Date (from‐to) : 2008 -2010 
    Author : ABE Koji
     
    For the research for enhancing a CBIR (Content-Based Image Retrieval) system of trademark images considering human subjects, the following subjects were realized. And then, the possibility they can be applied to CBIR systems for other images was investigated. (1) Models for the grouping factors of "proximity", "similarity", "closure", and "good continuity" in the Gestalt Psychology were proposed and a method for recognizing grouping areas in abstract trademarks by discrimination analysis was designed. (2) According to results of questionnaire on the image grouping using abstract trademarks, a model of a grouping factor "parallerism" was designed. (3) For the purpose of extracting metadata from sports broadcasted videos, using the phenomenon that the score banner in the broadcasted videos regularly change in every inning, all the flames in a inning are regarded as a group, i.e., a method for grouping flames into sets of flames in every inning was proposed by introducing the method proposed in (1). (4) For designing a computer-aided diagnosis in pneumoconiosis, a part of the models proposed in (1) was applied for noise reductions in chest radiographs obtained from a CCD scanner.
  • 日本学術振興会：科学研究費助成事業
    Date (from‐to) : 2006 -2007 
    Author : 阿部 孝司
     
    本研究で明らかにしたことは,以下の機能の実現である.これらの機能を申請者がこれまで開発してきた類似画像検索システムの機能として提案した. (1)抽象図形を対象に,ゲシュタルト心理学で明らかにされている図形の群化要因「近接」「類同」「閉合」「よい連続」の4つを測定するモデルを提案し,判別分析により図形内に存在する複数の群化領域を認識する手法を開発した. (2)抽象図形を対象に,複数名から採取した図形群化に関するアンケート結果より,ゲシュタルト心理学で明らかにされている図形の群化要因「平行性による類同」を測定するモデルを開発した. (3)スポーツ映像のメタ情報を抽出することを目的として,放送型野球中継映像を対象に,画像内のスコアテロップがイニングごとに規則的に変化することを利用し,同じイニング内のフレームを同一とみなす,つまり,1イニングごとで群化されるフレーム集合をクラスタリングする手法を開発し,(1)の動画像への適用可能性を検討した. (4)マンモグラフィ(乳房X線像)を対象に,画像内に点在する石灰化領域,また,それらが群化現象によりまとまって一つの領域に知覚される石灰化クラスタを検知する手法を提案し,(1)の医用画像への適用可能性を検討した. (5)風景写真(カラー画像)を対象に,ユーザが簡易な検索キーをラフペインティングして作成し,風景画像内の微妙に色や濃度が変化するグラデーション領域の全画素を同一画素値とみなして群化させることで検索キーとマッチングを行うインタラクティブな画像検索システムを試作し,(1)のカラー画像への適用可能性を検討した.
  • 日本学術振興会：科学研究費助成事業
    Date (from‐to) : 2004 -2005 
    Author : 阿部 孝司
     
    人間は,視野中に複数の小図形が塊状に点在して表現されているとき,それらを一つのまとまりとして知覚統合する.視覚心理学ではこの現象を群化といい,その要因はゲシュタルトの心理学により示されている.図形には,図形全体あるいはその一部の領域が群化していると認知できるものも多く存在するが,人間が図形内の群化領域を知覚統合した上で互いに類似していると判断した一組の図形に対し,計算機で群化領域を認識することは難しくこれらが類似していると判断できないことが多い.そこで,抽象図形内に存在する群化領域の認識を検討し,以下の成果を発表した. 1.人間の知覚群化の要因を定量化するモデルを作成した. 2.図形の群化領域の知覚に関するアンケート結果を実施し,人間は主観的に複数の群化要因にそれぞれ重みを付け,見方を変えて1つの図形から複数の群化パターンを知覚することを示した. 3.1と2を基に図形内の局所領域を統合していき群化領域を認識する方法を提案した. 4.提案した群化領域の認識手法を類似画検索システムに適用し,その有効性を示した. 5.提案手法をカラー画像検索や医用画像解析への適用を検討した. また,以下の検討点・発展性がわかり,これらの解決を今後の検討課題としたい. (1)人間は,図形内の局所領域から徐々に周囲の領域に視野を拡大させていき最終的に大局的に群化領域を知覚する場合と,初めに図形全体を観察し図形内容を認知した上で局所領域を個別に知覚する場合が存在することがわかり,現時点では前者のみの検討に留まっている. (2)群化要因の一つである「連続性」に関して,人間は(1)で示した後者に相当する知覚を行っていることが考えられ,3.の考え方では「連続性」を定量化するモデルを作成することは難しい. (3)CV,医用画像解析に関しても類似画検索と同様に,人間の知覚群化を考慮することで計算機システムの精度が向上する.

Other link

researchmap


Copyright © MEDIA FUSION Co.,Ltd. All rights reserved.