Study On Several Key Technologies Of Diagnostic Information Extraction For Face Diagnosis In Traditional Chinese Medicine

Face diagnosis is an important part of the inspection in Traditional Chinese Medicine (TCM). It's a necessary component in clinical diagnosis. TCM considers the body of human as an inseparable whole and the face is just like a mirror that reflects physiological function and pathological changes. The pathological changes of viscera can be directly diagnosed by inspecting the changes of complexion and eye expression.The beauty of face diagnosis lies in its simplicity and immediacy: whenever there is a complex disorder full of contradictions, examination of the face instantly clarifies the main pathological process. Therefore, it's of great value in both clinic applications and self-diagnosis. Moreover, Face diagnosis is one of the few diagnostic techniques that accord with the most promising direction in the 21st century: no pain and no injury.Traditional face diagnosis has inevitable limitations that impede its medical applications. First, the clinical competence of face diagnosis is determined by the experience and knowledge of the practitioners. Second, Face diagnosis is usually based on the detailed visual discrimination. Therefore, it depends on the subjective analysis of the examiners, so that the diagnostic results may be unreliable and inconsistent. These disadvantages bring difficulties to the further development of face diagnosis. Therefore, the objective research of face diagnosis is of great significance to the dialectical standardization, teaching and research methods of modernization of TCM.In this theis, the objectification of face diagnosis is specifically referred to the automatic recognition, quantification and dialectical deduction of two main facial diagnostic informations: complexion and eye expression. This research was to fill the blank on the technologies for automatically extracting facial diagnostic information in TCM. More specifically, this research aimed at building several computational methods for automatically extracting the information of complexion and eye movements. And the methods will provide theoretical and methodological support for the quantification, analysis and processing on the diagnostic information of face diagnosis in TCM. The contents of this research are several key technologies for automatically extracting the information of complexion and eye movements, including multi-view face detection, facial feature localization, complexion recognition in TCM and face diagnosis oriented eye tracking.The main research results were summarized as follows:Firstly, a method for multi-view face detection under complex background was proposed. The corners were utilized to directly extract candidate face regions from images, and consequently, the decline in detection rate caused by false estimations on the inclination of faces was avoided. Besides, two revisions to the detection procedure are as follows: (a) accelerate object detection by applying a rule of image edges; (b) utilize the L component of LAB color model to correct light variations in the images. The experiments on CMU frontal face and rotated face test sets result detection rates of 95.1% and 94.6% with 43 and 24 false alarms respectively. The experiment on a Feret profile test set result an 89.7% detection rate with 15 false alarms. Results show that the method not only realizes the detection of frontal, plane rotated and profile faces, but also has high robustness to occluding, light variations, image resolutions, etc. In addition, the experiment on the face database for face diagnosis in TCM results a 100% detection rate with 0 false alarms, which means that the method can be used to process and analyze the images for face diagnosis in TCM.Secondly, a method named 'FC-ASM' for object contours extraction was proposed. First, the results of FCM clustering on images were taken as the initial position of C-V segmentation model, and consequently, the convergence of C-V model was accelerated. Second, the classic ASM matching procedure was revised by taking the partial contours that extracted by C-V model as fix points. The object's geometric information and statistical model's prior knowledge were fully utilized. Exact segmentation was made to the regions that have relative strong geometric information. And for the regions that have relative weak geometric information, statistical knowledge was utilized to extract contours purposefully. The method was successfully implemented to facial feature contour extraction, and the accuracy surpasses AAM which is the mainstream method for facial feature localization by 27.2%. The accuracy and robustness of the method are so good that it can be used to provide accurate reference positions for automatically extracting complexion in TCM.Thirdly, based on complexion-viscera diagram in TCM, a method for complexion recognition in TCM was proposed for the first time which achieves an accuracy of 84.6%. First, the environment of shoot was standardized. Second, the complexion features in LAB color model were extracted from the corresponding feature points on complexion-viscera diagram by FCM clustering which separates the complexion from skin color. Third, the complexion features were classified by SVM automatically. Results of the method not only can be served as a basis for the automatic logic inference of face diagnosis, but also have reference value for clinical diagnosis.Finally, a face diagnosis oriented eye tracking model was constructed. First, the lighting condition was standardized. Second, the nostrils were localized and taken as reference points, and the left eye's relative movements to the nostrils were taken as the real movements of the eyes. Third, the Camshift algorithm and Lucas-Kanade optical flow algorithm were respectively utilized to track the face and nostrils real-timely. Finally, the rate and trajectory of eye movements were calculated. The experiments on a video of resolution at 640*480 result a tracking speed of 25 frames per second. Results of the model lay a foundation for eye expression analysis in TCM.All the experimental results proved the efficiency of the proposed methods, and the goal of this research was mainly achieved. In sum, this research has significantly scientific study value that will expand the study on the objectification of Four Diagnosis and will also enrich the technology for extracting diagnostic information in TCM. In addition, face related technologies, such as face detection, face recognition, facial feature localization, face tracking and so on, are research hotspots in the field of pattern recognition for the difficulties in realization and their wide application. So, the research results could also be the reference for other related application researchs.