Performance Comparison between Color and Spatial Segmentation for Image Retrieval and Its Parallel System Implementation

In this paper segmentation on both color and spatial space is implemented to compare their influence to the performance of content-based image retrieval (CBIR) system. Firstly, images are converted from RGB color space to HSV space. And then cumulative histograms on H component are extracted from four segmented blocks as one color feature of the image. At the same time, the HSV color space is further divided into H2SV space. Local accumulative histograms on H1 and H2 components are sequentially extracted as the other color feature of the image. Finally, both color features combined with the texture feature are taken to evaluate their performance of retrieving images. In addition, some parallel techniques are applied to construct an image retrieval system based on the cluster architecture. Experimental results have demonstrated that spatial segmentation has done more influence on the performance of CBIR system. And the parallel techniques can improve the retrieval efficiency significantly.