Method And System On Facial Expression Recognition And Intelligent Tongue Diagnosis Based On Convolutional Neural Network

Facial images contain a lot of information related to the human body state.Effective detection and recognition based on facial images improve practicality and intelligence of image processing.For example,expressions are associated with psychological state and tongue features reflect physiological health state.Image processing technology is widely used in artificial intelligence fields such as clinical psychology,health diagnosis and automotive safety.Expression recognition and tongue diagnosis are both image processing technology based on facial images,and the recognition process is consistent,mainly including three steps: image pre-processing,feature extraction and classification.Facial expression recognition is currently dominated by deep learning methods.Convolutional Neural Networks are constructed for classification.Deep networks extract complex features,however,with large computation and slow operation,applications in low power platforms such as embedded devices are limited.Tongue diagnosis is mainly based on traditional artificial methods.It has large workload and is dependent on subjective experiences.Therefore,it is necessary to extract and learn tongue features by deep learning methods.A model with high performance is built to classify human constitutions,thus the intelligent tongue diagnosis is realized.Improvement methods are proposed from detection,recognition and system to address the problems of low sample quality,poor model performance and limited practical application in the study of expression recognition and tongue diagnosis.The main work are as follows:(1)Pre-processing operations such as data enhancement and grayscale are performed to improve image quality for feature extraction and classification.A face detector based on Haar and Ada Boost is constructed and visualization function is implemented using Open CV.Detection results indicate the proposed method has good robustness for occlusion and deflection.(2)A shallow-CNN based on Xception,depth-separable residual module and SE block is proposed.Experimental results on CK+ and FER2013 datasets show that the proposed model with fast running speed and high recognition accuracy is more competitive than many popular models.An expression recognition system is built,which consists of modules such as face detection,model recognition,soft labels and expression intensity classification.The system recognize facial expression and expression intensity level effectively.(3)A tongue detector based on HOG features and SVM classifier is built by Open CV.VGG16 is improved in structure and parameters for tongue detection and TCM physique recognition.An intelligent tongue diagnosis system is developed which consists of health testing and health care modules.The system performs effective tongue detection and constitution recognition,and then intelligently matches corresponding health care program,realizing tongue diagnosis.