Technology Research, Based On Information Fusion Of Multi-feature Image Retrieval

Content Based Image Retrieval (CBIR) has been a very active research area since 1990's. In the content based image retrieval, it is retrieved by single visual character at the beginning of the research, such as: color, texture and figure characters. Color and texture character are two very important and widely used image characters in the content based image retrieval. So firstly these two features are respectively discussed in this paper. And at the same time invariable theory has achieved successful application in the content based image data retrieval in recent years, so invariant character of image retrieval is also discussed in this paper.First, two color spaces (RGB and HSV), three color hsitograms methods (tradition hsitogram, accumulation hsitogram, local accumulation hsitogram), and color matching algorithm are discussed in this paper. Then texture feature is analyzed and discussed,and gabor wavelet transformation is used to pick up texture feature. Afterwards,this paper briefly describes related content of invariant theory,and chooses the invariant descriptor of combining with DFT and log polar transformation to be image character descriptor, and does the relevant image retrieval experiment.Relevance feedback technique has been an important approach in image retrieval discussed in this paper. Meanwhile this paper quotes a novel image retrieval method of support vector machine based relevance feedback, and the proposed approach partially solves the sample insufficiency problem that exists in SVM based relevance feedback algorithm, and show the approach has good performance for retrieval.The image retrieval fusing with multi-character is a progress direction. However how to efficiently fuse and affirm the weight value among different characters in the traditional relevance feedback is a big problem. This paper, with the idea of information fusion, proposes a two-layer SVM classifier model to improve the precision of the classification and uses SVM-based Information Fusion Machine to efficiently solve the problem of fusing multi-character in the image retrieval. It radically solves the big problem of distilling and fusing multi-character. Many experiments prove that the three image characters, that is, invariant character, color and texture character, use the fusion method well. Furthermore, the effect of image retrieval improves a lot comparing with that of the traditional method.