Graph-based color image segmentation using edge magnitude for image understanding

摘要

This paper proposes color image segmentation algorithm which is aimed to assist image understanding. The proposed method is based on graph-based approach and employs comparison function proposed by P.F. Felzenszwalb et. al.. At first, variable weighted gaussian smoothing filter, which is also proposed in this paper to remain edge property, is performed to an image. Next, graph is generated, in which eight neighbor pixels are defined as nodes and, the distance of color in L{sup}*u{sup}*v{sup} color space and edge magnitude are defined as edge weights. Based on this graph, image is segmented. Finally, the segmented regions are integrated using inclusion relation among regions.