Image segmentation through index images

Image segmentation is an important pre-processing step in the applications of MPEG-4 and computer vision. It is one of major challenges in MPEG4, since MPEG-4 is constrained by how well previous segmentation is accomplished. In this paper, we propose a scheme called SII (Segmentation through Index Images) algorithm for image segmentation. We use the histograms of index images as the features to classify the image data into several clusters for segmentation. The 1 st-moment index image is the low-resolution image that its pixels value which is the index of codevector to label all corresponding image blocks. Through the 2nd-moment index image, we identify image blocks which belong to contour or homogeneous blocks. Then, we can connect all homogeneous blocks which have the same label to define an interior of a region. The contour block is split into the smaller size and assigned to one of its neighboring regions to obtain the more accurate contour of objects. We present several experiments to show the efficiency of our STI algorithm. According to simulation results, our SII algorithm is effective for image segmentation.