Abstract: An improved fast random walk algorithm is proposed to solve the problems that traditional random walk algorithm only utilizes grayscale information so that it is not able to segment color tongue images accurately and its efficiency is relatively low.Firstly,the original image is segmented by the toboggan algorithm with compression rule.Secondly,a novel compound weight function is introduced and a weighted-graph is built and simplified.Thirdly,random walk algorithm is adopted to make clustering to the weighted-graph.In the end,mathematical morphology operators are used to repair the target region and get the target tongue image.As the experimental results show,the improved random walk algorithm achieves ideal segmentation effects and its efficiency is much higher than those of the former algorithm and the traditional random walk algorithm.