Image Segmentation based on Geodesic Distance Combined with Region and Edge Gradient
Xiangyang Liu, Yong YuJie, Xiaofeng Zhang · 2020
In this paper, data points are uesd to represent vertices of a graph connected by weighted edges signifying similarity based on distance. The goal of this work is to accurately achieve image segmentation using both edge and region information on the basis of superpixels. In order to combine these clues in an optimal way, we use geodesic distance and formulate features that respond to characteristic changes in brightness, color, texture and gradient associated with adjacent superpixels. To determine whether a superpixel boundary is a segmentation, the edge gradient and the local density are used to overcome some shortcomings of the traditional DP algorithm. Experiments are conducted on Berkeley Segmentation Database and the result of segmented images verify the efficiency of our approach by comparison with other methods.