Diagnostic Information Extraction For Face Diagnosis In Traditional Chinese Medicine

For sake of the Objectification of Four Diagnosis (Observation) in Chinese Traditional Medicine, diagnosis information extraction and processing mainly including the pulse condition and the tongue information extraction are studied and the processing technology already launches. In the latest five years, the tongue diagnosis information processing has developed enormously, and clinical practice has already started. Relatively, as the same important part of TCM diagnosis, looks information extraction and procession are still at the research stage of blank. In fact, concerned with the diagnosis of tongue information, looks information play an irreplaceable role. Studying on the key algorithms about head looks information extraction is extremely urgent with important significance.In this paper, we probed studying various image processing algorithms which are suitable for Traditional Chinese Medicine diagnosing by looking at the color of face. The purpose is to find the most effective algorithm on such image processing as face detection, feature points localization, color information extraction and face color belonging. Our main work and innovation in this paper lies:1. On face detection, we mainly studied the face detection technology based on Guass skin model and Adaboost Classification. First of all, construct Guass skin model and get the rough position of face through this model, then construct Adaboost face Classification and detect face by this Classification, finally according to the detection result of the two methods, we can get the face position after analysis of the relative position. Test shows, combination of the two algorithms guarantee an accurate rate of 100% in face detection when the given image is taken on the standard illumination condition.2. On localization of feature points, object detection based on Adaboost Classification and feature points localization based on Active Shape Model (ASM) are mainly studied. In the very beginning, train eye, nose and mouth Classifier respectively like training the Adaboost face classifier in face detection. Given an image with a face on, load the Classifiers and detect eye, nose and mouth respectively, then on the help of the connatural relative position of which, we can easily exclude the improper detection and get the rough position of feature points on the face, which are used as the beginning shape in feature points localization based on ASM. So through the object detection based on Adaboost classification, ASM is used for the accurate positing of feature points, and then we get a higher accurate rate.3. On color information extraction, we correct the color based on the basic skin-color(generally the color of the skin at the arm near the hand is considered to be the basic color ), and then compute color value mean in different Color Mode according regions which is segmented by feature points on face. On face color belonging and recognizing, we studied SVM and used it in face color's classification creatively, and also get achievements in this stage.