Deep Learning Based TCM Tongue Image Color Correction,Segmentation And Server System

After years of fundamental research,the objectification of Traditionaal Chinese Medicine(TCM)tongue diagnosis has achieved greatprogress in some extent.In the productization process of TCM tongue image analysis system,there are still many chanllenges,such as the color reproduction of the subjective and objective tongue images is not coordinated,and the interactive-free tongue image segmentation.Accuracy and robustness of segmentation are difficult to meet the requirement of fully automated with interactive free.These challenges have largely limited the productization of the automated analysis technology of TCM tongue images.In recent years,with the development of deep learning in the field of computer vision,it is widely used in various regression problems and image semantic segmentation tasks.This provides an opportunity to solve the above problems in the productization of TCM tongue image analyzer.In this thesis,we have investigated on the subjective and objective color reproduction of TCM tongue images,the automatic segmentation of tongue images and the algorithmic call framework of deep learning algorithms in TCM tongue image analyzers.The main contents of thisi thesis include the following sections:(1)A two-step deep learning based subjective and objective color correction method for TCM tongue images.The method formulat the color correction of TCM tongue images in two steps,which meets the requirements of the objective quantification of automatic analysis and the subjective individualization of human visual system.The collected tongue imagesare objectively corrected by the objective correction stepof TCM tongue image based on convolutional neural network.Then,according to the subjective preference of different doctors and the statistical analysis results of different environments,the color adjustment based on l?? color space is adopted.The strategy adjusts the subjective display of the tongue images.The experimental results show that the objective error of the algorithm is greatly reduced compared with the traditional method,and it has the flexibility to meet the subjective individualization requirements.(2)An automatic segmentation method for TCM tongue images based on convolutional neural network is proposed.The method employed the semantic segmentation framework of codec,and uses the tongue image dataset in the open environment to carry out transfer learning on the semantic segmentation network,and obtains the tongue segmentation model adapted to the automatic segmentation task of TCM tongue image.This method can be applied to tongue image segmentation in closed and open acquisition environments.Experimental results show that the method can realize the automatic and accurate segmentation of the tongue images,and achieves the practical level.(3)The framework of the deep learning model of TCM tongue diagnosis based on Web server is constructed.The method adopts the lightweight Flask web framework to implementthe convenient integration of the deep learning algorithm on the server.At the same time,it completes the application development of the traditional Chinese medicine tongue-like APP based on Android,and implements the mobile application end to the Chinese medicine tongue diagnosis deep learning model call framework.The experimental results show that the Chinese medicine tongue deep learning model call framework effectively solves the integration engineering problem of different platform deep learning algorithms in the traditional Chinese medicine tongue image analysis system.