Image retrieval system based on multi-feature fusion and relevance feedback

The content based image retrieval system may discovery user needed image fleetly and accurately from image database. In order to improve content based image retrieval system performance, this paper introduces a new method which realizes image retrieval by multi-feature fusion. Color feature is extracted based on color histogram, texture feature is extracted based on gray co-occurrence matrix, and shape feature is represented by moment invariants. In accomplished image retrieval system, three kinds of low-level visual features of image are fused correctly, and relevance feedback and weight regulation algorithm are utilized to increase the precision of image retrieval. The feature extraction, system structure, similarity calculation, and working flow are also discussed in detail. The experimental result indicates that multi-feature fusion can improve the accuracy rate and recall ratio of image retrieval, and increase system efficiency, validity, and flexibility.