Image Retrieval Based On the Color-Spatial Distribution Feature of Generalized Images

A new image indexing based on the color-spatial distribution feature is presented in this paper. The generalized image model that combining the original image with its smoothed image be used in the retrieval algorithm. In order to describe the spatial distribution of color, the annular color segmentation algorithm is introduced. According to the annular color segmentation algorithm, equidistance annular color-spatial moments be calculated as the features of color that can denote the color-spatial information of the generalized images to a certain extent. However the annular color segmentation algorithm can not distinguish the visual attention of the different segmentation region. Therefore we design the weight for different segmentation regions that can express the visual attention of the different segmentation region in the retrieval algorithm. Thus the weighted color-spatial distribution moments can be used to describe the color-spatial feature of generalized image and consistent with human vision. Finally, the character vectors are normalized using Gaussian model and L1-norm distance be used to measure the similarity between the different images. Experimental results show that the proposed algorithm in this paper outperforms Geostatic algorithm in the color image retrieval.