Automatic Classification of Tongueprints in Healthy and Unhealthy Individuals

Abstract

Tongueprints are fissile texture on tongue and one of observational contents of tongue diagnosis, which is an important diagnostic method in Traditional Chinese Medicine (TCM). With deep researches on tongueprints, three basic problems emerge: (1) TCM has always held that healthy individuals do not exhibit tongueprints, but in the recent years some medical researchers have found some healthy individuals in their small sample (< 1000 cases) tongue image database do exhibit tongueprints. How about do tongueprints in a large amount of healthy individuals (> 2000 cases)? (2) If about a third of a large amount of healthy individuals have tongueprints, mainstream diagnosis by inspecting tongueprints should lead to an over-diagnosis problem and some healthy individuals are diagnosed mistakenly as unhealthy individuals because it holds tongueprints themselves declare publicly diseases. Thus it is necessary to diagnose definitely a tongueprint image belong to an unhealthy individual based on tongueprint features. This is a basic problem and is of theoretical and practical importance for diagnosis by inspecting tongueprints. And which features of tongueprints can be used to diagnose definitely a tongueprint image belong to an unhealthy individual? (3) Actually, the second problem is recognition and classification of tongueprints in healthy and unhealthy individuals. To further promote the modernization process of the traditional tongue diagnosis Chow are the researches done on automatic classification of tongueprints in healthy and unhealthy individuals? After further making sure there appear tongueprints in healthy individuals and finding which features of tongueprints can be used to diagnose definitely a tongueprint image belong to an unhealthy individual by statistic analysis on a large database of tongue images (more than 3000 cases), this paper do the researches about automatic classification of tongueprints in healthy and unhealthy individuals based on the large database. Firstly tongeprint regions are extracted by multi-direction synthetic tongueprint extraction. Then a SVM classifier is used to recognize a tongueprint image belong to a healthy or unhealthy individual based on tongueprint features. Recognition accurate rate of tongueprints in healthy and unhealthy individuals respectively are 75.81% and 74.61% based on the large database.