Segmentation based on shape prior and graph model optimization

Abstract

A scheme of segmentation based on low-level and high-level cues is presented. Firstly, image-pyramid is obtained based on segmentation by Weighted Aggregation (SWA), the suitable coarse pixel image is selected to be as low-level segmentation cues. Kernel principal component analysis (KPCA) is used for building the space of shape to represent shape prior knowledge. The coarse pixel image is expressed through a graph model, based on high and low level cues, genetic algorithm (GA) is used to find out the optimal sub-graph to segment object precisely. Experimental results demonstrate that our proposed approach is able to accurately segment the objects with better performance than the existing methods.