Study Of Teeth-marked Tongue Based On Image Processing And Pattern Recognition

In modern society, irregular diet and sleep habits and multiple pressure from outside are a serious threat to human health. TCM tongue diagnosis can effectively predict the function of human viscera and Qi and blood, but it relies on visual observation and it restricted the development of TCM tongue diagnosis. So, the modernization, objectification and standardization of TCM tongue diagnosis is a problem to be solved.This research proposes some improvements for the existing teeth-marked tongue analysis algorithms and some new analysis methods. In paper, firstly some preprocess measures are conducted including coarse segmentation based on gray-level projection, PGF detection and ADW VMF filtering, luminance component contrast enhancement, tongue extraction based on complementary components merge and teeth-marked tongue judgment by concavity calculation; secondly the extraction of teeth marks is done by establishing edge depth curve, edge R value curve and edge I value curve, of which moving average method, Douglas-poke method and polygonal approximation method are used to smooth curve; thirdly, maximum, minimum, mean and variance of the depth, length and area of teeth marks are extracted as new features; fourthly, the classification performance of features are studied with LIBSVM and10times5-fold cross validation. Besides, the system acquisition scheme and angle identification algorithms are studied, and the lighting environment and mechanical structure of tongue image acquisition device are designed.Experiment results show that improved number extraction method can better solve the fitting precision problem; new methods based on edge color change and gloss change can extract number of teeth marks of teeth-marked tongue without obvious edge change. Besides, the number-depth-area combination feature has better classification performance than single number feature, which reaches73.39%.