Tongue Feature Extraction And Research Of Fusion Classification

Tongue diagnosis is one of the most valuable, popular and widely applied diagnostic methods of Traditional Chinese Medicine. The research of"Automation of TCM Tongue Diagnosis"has great significance to the whole society.From the information fusion point of view, in application to the principle of Feature Layer Fusion, this dissertation designs an AdaBoost classifier based on CART, combines the features of tongue images obtained by using different methods into one vector, let the vector be the input of classifier, chooses the feature during the train process and get the result finally. The whole process can be divided into two parts: feature extraction and classifier design and realization. Technology of information fusion, image processing, pattern classifier and so on are applied.Information extraction can be divided into two parts: lingual information extraction and sublingual information extraction. In the former part, the color and texture features of tongue surface are extracted. In order to get the most comprehensive information, color feature is described and extracted in RGB, XYZ, CIE-Lab and CIE-LUV color space and texture feature is quantized by applying both Gabor wavelet and Gray level co-occurrence matrix. For the part of sublingual information extraction, this dissertation firstly takes the sublingual vein from the near infrared sublingual images in to consideration. Apply mathematical morphology methods such as watershed, dynamic thresholding and iterative erosion with variable structure to acquire the entire contour of the sublingual region, remove the glisten points and other noises, complete the binarization of the obtained effective sublingual region, and further confirm the candidate sublingual veins regions. Finally, region growing is iteratively performed to trace the contour of sublingual veins.The AdaBoost algorithm based on CART decision tree is designed as the classifier in this dissertation. Color and texture features are taken as the input data and the judgment of whether the image under consideration taken from a healthy or an unhealthy person is given as the final result. Moreover, analysis and testing on the feature selection ability of AdaBoost are implemented, and the experimental results prove that the designed AdaBoost possesses great performance on feature selection. The final experimental result shows that the algorithm can obtain a favorable effect on classification.