Computer Aided Tongue Diagnosis Classification Models Based On Color Features

Tongue diagnosis is a vital basis of Traditional Chinese Medicine (TCM)diagnosis thanks to its simple and noninvasive characteristic. The result of diagnosis isinfluenced by the doctor’s knowledge and clinical experience. Moreover, someexternal factors can also make poor diagnosis, such as the illumination and angle. Thepromotion and development of tongue diagnosis are limited by these factors.To address these issues, computer aided tongue diagnosis has been developedby combining tongue diagnosis with computer image processing technology. Themain process of this new technique is organized by sampling, segmentation, featureextraction, machine learning and other steps to make objective and accuratereflection on patients’ pathological information, reduce the negative impact of otherfactors, assist physicians to find potential diseases, improve the accuracy ofdiagnostic results, and implement tongue diagnosis to be more objective andquantitative.The feature of color is a connection between tongue diagnosis and imageprocessing. It is not only the primary consideration when observing the patient’soverall status by using tongue diagnosis, also being the most common information usedin image processing. Significant previous research has been done based on the tonguediagnosis analysis, but they have a common problem—selecting the features blindly.In this thesis, I have done a series of studies based on previous work. First, Idesigned a new class of color feature, reduced its dimensionality by using SVMsupport vector machine(SVM) feature selection method and improved theclassification performance. Secondly, adopted the SMOTE over-sample technique inorder to obtain balanced samples, and solved the problem of imbalanced samples inthe application. Finally, the experiment has been extended into3-state classification in addition to the binary one, and obtained promising results. All the above has laid agood foundation for further study.