Technology Research, Content-based Visual Information Retrieval

As one of most important vision information of an image, the calculation of color characteristics is simple and steady and this has been broadly used in the Content Based Image Retrieval (CBIR). The expression of image color, the extraction of the color characteristic and the matching algorithms of image likeness based on colors are analyzed and discussed in this paper. First, this thesis introduces the comparison to several common color spaces. On these basis, three matching algorithms of color histogram (histogram intersection, the distance of Euclid and proportion distance) and three color-histogram methods(tradition histogram, accumulation histogram, local accumulation histogram) are discussed in this thesis, the retrieval effects are compared through the experiments.In this paper, we develop a prototype systems started with the accuracy and time of the content-based image. Relevance Feedback and Support Vector Machine learning are adopted in the contend-based image retrieval system. We also compare DTW&GSVM results to LJFM&GSVM's which proves that the algorithm of DTW&GSVM is effective.There are 5 sections in this paper. In section 1, we describe the research state of content-based image and jobs we have done. In section 2, we introduce image retrieval algorithms based on color features and histograms. The design and realization of a space relationship-based image retrieval prototype system is provided in section 3. And finally, we conclude this thesis in section 4, together with a discussion of future work.