Research On Image Segmentation And Recognition Of Tongue Diagnosis In Traditional Chinese Medicine

Tongue diagnosis is one of the important diagnostic methods in traditional Chinese medicine.In recent years,with the development of image processing and machine learning technology,the study of computerized tongue diagnosis system has been paid more and more attention.A computerized tongue diagnosis system contains three parts: tongue image acquisition,tongue body segmentation and tongue manifestation recognition.Image processing and machine learning technology is used to identify and classify the tongue information,and finally the diagnosis result is obtained.In this paper,we will study tongue segmentation and tongue manifestation recognition methods that do not depend on the acquisition conditions,considering that the tongue diagnosis system should be applied to various platforms in order to promote TCM diagnosis technology.The work of this paper includes two parts of tongue body segmentation and tongue manifestation recognition:Tongue segmentation is the first step and key content of computerized tongue diagnosis.In this paper,we propose a method for segmenting the tongue through coarse to precise segmentation.The first step is to conduct a coarse positioning of the tongue.Firstly,the skin color detection algorithm is used to remove the complex background.Then the H component is shifted in the HSV color space,and the HSV components are smoothed by the mean shift algorithm.Finally,the rough location results are obtained in L*a*b* color space.The second step is to achieve accurate segmentation by improving the marker-controlled watershed algorithm.Firstly,the foreground marker is obtained by morphological technique.Combining the foreground marker with the localization result,a new foreground marker is obtained.Then the watershed algorithm is adopted to segment the image.Finally,the final accurate segmentation result is obtained by the geodesic active contour model.In the part of tongue manifestation recognition,this paper presents a method of recognizing and extracting the tongue manifestation of tongue spots or petechiae.The first step is recognizing the tongue manifestation of tongue spots or petechiae.Firstly,the blob detection algorithm is used to detect the blobs.Then the eigenvectors such as the number,size and distribution of the blobs are calculated,and the eigenvectors are generated.Finally,the support vector machine is used to get the classification model.The second step is the extracting of spots and petechiae.Firstly,the blob detection parameters are changed to get the second detection result,and then the spot color feature is extracted.Then,the spot detection results are clustered into several small clusters using k-means clustering.Finally,by defining the discriminant function based on the weighted color space distance,the above clustering results are compared with the result of the first spot detection,and the extraction result is obtained.Experimental results show that the tongue segmentation algorithm proposed in this paper has a good segmentation effect on the tongue image under various acquisition conditions.In the part of tongue manifestation recognition,the accuracy of spots and petechiae recognition is 97.4%.The false detection rate of the extraction is 6.0% and the leakage detective rate is 10.1%,which show the effectiveness and application prospect of the method.