Research On The Techniques Of Region-Based Image Retrieval

Region-based image retrieval (RBIR) is an important research branch of CBIR. Region-based retrieval applies image segmentation to decompose an image into regions, and uses region visual feature to represent and index the image. It can reduce the gap between low-feature and high-level semantic concepts and more close to the human perception.The paper analyzed and summarized the fundamental, key techniques and performance evaluation of CBIR. We mainly developed and discussed the retrieval on common color images. An algorithm based on color and spatial information is proposed in this paper. First, k-mean cluster is performed on an image based on color coarseness metric. Second, the merge / segment operation is performed based on cluster's color simplicity, pixel's adjacent relations and region's space station. The paper test the performance of color means, 256 bin color histogram and 72 bin histogram, and two multiple-feature distance measure. In the prototype system, 256 bin color histogram and moment were chosen as region feature, the distance between two regions was defined by the expression proposed by SIMPLIcity, and the IRM measure was used to calculate the distance between two images. A method is proposed to adjust region importance in IRM.