Development of CBIR System Considering Perception of Grouping Areas and its Applications

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K00258
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Perceptual information processing
• Research Institution: Kindai University
• Principal Investigator: ABE Koji 近畿大学, 理工学部, 准教授 (90367441)
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: 類似画像検索 / CBIR / ゲシュタルト心理学 / 群化 / 画像認識

Outline of Final Research Achievements

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.

Free Research Field

メディア情報学

Academic Significance and Societal Importance of the Research Achievements

本研究により、内容ベースの画像検索の性能が向上しSemanticな画像解析へ応用できることを示した。すなわち人間の知覚機能を導入した類似画像検索システムを構築することに近づき学術的意義のあることが示唆される。また、画像認識に群化認識手法を取り入れることで種々の画像認識を行う社会システムの精度が向上したことを示し社会的にも意義のあることが伺える。また、昨今画像認識で用いられる深層学習では学習データの量が精度に大きく依存しビックデータを必要とするが、大量データを取得できないケースも存在する。本手法ではそのようなケース（医用画像処理）で有効に機能したことからも学術的な意義は十分にあることが伺える。