A Salient Region of Interest Based Retrieval Scheme for Image Databases

This paper proposes an image retrieval system scheme based on salient region of interest. In our retrieval system, an image is represented by salient region of interest and background. The salient regions of interest roughly correspond to semantic objects, which separated from the background by a robust automated segmentation method based on the edge and region integration. The measure for the similarity between images incorporates properties of two categories of regions and the entire image. Experiments show that our scheme can improve the quality of retrieval compared with global similarities based query.