Image Retrieval Based On Roi With Grid And Quotient Space Multi-granularity Theory

With the rapid development of computer and Internet, Content-based Image Retrieval (CBIR) becomes one of the most active research focuses of multimedia. It is an interdisciplinary research area of artificial intelligence (AI), computer graphics, pattern recognition, psychology, and etc. It extracts some features from images, finds the similar images with the request image in the features space and finally carries out images retrieval, inquiry and management automatically and intelligently.The main problem that exists in CBIR is the "Semantic Gap" between the low-level features and high-level semantic concepts. That is the low-level features can't reflect the real semantic concepts. To overcome this problem, people have proposed some methods recently. Such as retrieval based on region of interesting (ROI), relevant feedback and classification model. This paper focuses on the active research—image retrieval based on ROI and proposes an image retrieval method based on ROI with Grid and Quotient space Multi-granularity Theory.Firstly, the background, the signification, the research status and some typical systems of CBIR are introduced. Some key techniques in the field are analyzed and discussed systematical. Such as low-level feature extraction algorithm (color, texture and shape), the similarity measurements and relevant feedback.Secondly, the quotient space granularity theory and the implement in image processing are introduced. Aiming at the "Semantic Gap", image retrieval based on ROI that combines grid and quotient space granularity theory is proposed. It integrates color, texture and shape features and implies relevant feedback to retrieve image. It has good human-computer interaction and can minify the "Semantic Gap".At last, an image retrieval system based on ROI is implemented in this paper. The system uses grid to divide image, extracts some low-level features under different granularity and stores them in the feature database, determines the ROI by user and return the result based on mixture theory of quotient space granularity. Testing results show that the system can achieve a high recall and precision rate.