Facial Disease Diagnosis Method Research Based On Color Image

Observing facial color is an important part of the inspection diagnosis in TCM. Doctors refer patients with facial features by observing the color change to determine the overall health status and the corresponding internal organs, and it provides us the basis information about the diagnosis and treatment. Chinese medicine:"color for the blood of the wing, side of the hash for the gas, blood changes, color changes, facial color should be the most expensive; it is observed by inspection the first color, observing the color will be too heavy on the face". This fully demonstrates the facial color diagnosis in Chinese medicine the importance of attending law.However, the traditional diagnostic method is mainly looking through the medical intuitive observing facial color, language describe and experience discrimination, the diagnosis was well received by the doctor's knowledge, thinking, and limited diagnostic skills, but also by light, temperature and other external objective conditions, the lack of constant and objective clinical evaluation criteria. Face consultation is the use of modern computer analysis of facial features automatically. However, the image signals of the face elements are instability in different environments, often making the color face images are inconsistent, and color correction is needed. In addition, the accuracy of face extraction and color feature extraction could be improved. In this paper, the solutions for these problems are given as following:(1) Automatically color checker extraction strategy: Before the color correction, we need to know the color checker's values, so need to extract the color value. For color checker characteristics, we propose two extraction methods: one is based on the location relationship between the extracted blocks of color in color card, and the other is based on color clustering and image segmentation.(2) Extract the face region of interest (ROI): The face images are captured by three generation acquisition equipment, previous algorithm has problem to accurately process the face ROI regions. To improve it, the largest between-class variance method is used for determining binary threshold and it greatly improves the segmentation accuracy.(3) Feature extraction algorithm: For the previous extraction features, space dimension is large and the accuracy of classification needs to be improved. So we propose two new feature extraction methods: 1D histogram and primary colors color histogram. Particular primary color histogram feature can solve the problem of large dimension.(4) Classifier design: The traditional K nearest neighbor and support vector machine classifiers in medical image processing are the presence of some defects. So we propose improved ideas, and have achieved good results.