Tongue Image Segmentation Based On Random Walk Algorithm

Tongue diagnosis is an important auxiliary means in the process of Chinese medical diagnosis.The doctor can get information about illness through the observation of patient’s tongue.Because the traditional diagnostic methods rely on doctors’ subjective judgment seriously and are easy to misdiagnose,there are more and more researches on automation and standardization of tongue diagnosis today and trying to establish a standardized and objective tongue diagnosis system.This system usually concludes two parts: hardware and software.For software,the study on segmentation of tongue image is important.This subject focused on the algorithm study of tongue image segmentation.Before the analysis of tongue feature,the first step is to segment tongue body from lips and peripheral skin.These parts have similar color with tongue body and will impact the feature analysis.The complete and accurate segmentation of tongue body is of great significance to further research.In this subject,random walk method was applied to segmentation of tongue image.This algorithm is based on graph theory that has better computation efficiency and accuracy.Large numbers of datas were tested and quantitative evaluation was done.The results showed good effect and the accuracy could reach 94.3%.To solve the problem that the traditional random walk algorithm depends on the artificial participation seriously,this subject proposed two improved methods which was based on location information and was based on feature information to realize the automatic selection of the initial seed points.Also,large numbers of datas were tested and the accuracy of both of the two improved methods could reach 74%.Finally,this subject consummated tongue diagnosis software system and integrated the improved random walk algorithm in the system.Hope to help clinical doctors realize automation and standardization of tongue diagnosis.