关键词: 图像分割, 半自动化, 随机游走

Abstract: A new approach of semiautomated image segmentation based on multi-label pixel was presented. First, a weighted graph of un-processing image was defined by Gaussian weighting function. Second, the pixels of different goal region were labeled differently in source image. An un-labeled pixel was randomly selected as the starting point. Then, random walking started from this location according to the defined weighted graph, calculating the probability from the start point to each pre-labeled pixels. A probability graph toward each unlabeled pixel was obtained, which represented all probabilities that each unlabeled pixel randomly reached each of pre-labeled pixels. The greatest probability was taken as its objective, therefore, a high-quality image segmentation could be obtained.

Key words: image segmentation, semi-automation, random walker