Research On Tongue Classification Based On Color Dictionary And Doublet Feature Optimization

In recent years,with the vigorous development of computer technology,using image processing and pattern recognition technology to make the tongue more intelligent.The automated tongue diagnosis system overcomes the shortcomings of traditional subjective and non-quantitative,which is of great significance and reference value for tongue diagnosis.However,a tongue diagnosis system is often affected by many factors,mainly in two aspects,namely,feature extraction methods and classification strategies.In view of the above two problems,this paper mainly does two aspects of work.In the aspect of feature extraction,this paper proposes a method based on color dictionary feature extraction and doublet feature optimization.In the classifier,we adopt efficient GBDT classifier.There are two key points in the color dictionary-based feature extraction method:The first is the definition of the color dictionary.Before the definition of the color dictionary in this paper,we first analyze the color gamut of the tongue in the CIELab color space,and find the main color of tongue image.Definite tongue color dictionary according to the main color of the tongue color in the CIELab color space.The second is the feature extraction,image segmentation technology and the pixel value similarity principle were used in the process of feature extraction.Similarity of between the image blocks and the color dictionaries is analyzed,and then the histogram features of the segmented blocks are extracted.Finally,the histogram features of the segmented blocks constitute the final tongue feature.As a result of the overlapping segmentation method,the integrity of the image information is maintained;and the local feature of the image is extracted,so that the image can maintain the invariance when the image is scaled.Doublet feature optimization method is used to optimize the extracted feature.Firstly,the doublet is constructed by using the similarity of sample,and its new sample features and categories are defined.Then,the polynomial kernel function model is used to define the new kernel function for the doublet.The model of the doublet is obtained by classifier training.The model reconstructs the original samples that have not been processed by Doublet,thus optimizing the original sample characteristics.In the classification strategy,we use GBDT classifi,er.GBDT is a combined classifier.It uses residual as loss function.The weak classifier uses CART,and then combines multiple weak classifiers into the final strong classifier.In order to prove the effectiveness of the method,the tongue data set is verified.The experimental results show that the feature extracted by the color dictionary feature extraction method has strong robustness,and the Doublet feature optimization method reduces the characteristic noise interference.GBDT is compared with the SVM that has a higher efficiency in classifying tongue images.