Facial Expression Recognition In Rhesus Monkeys Based On Facial Action Coding System

With the development of the times,the study of rhesus monkey facial expressions has made a significant contribution to medicine.For example,Parkinson’s disease,schizophrenia,autism and other diseases are accompanied by changes in facial expression.Studies have found that rhesus monkeys and humans have certain homology in facial expression.Therefore,in some clinical disease studies where facial expressions reflect disease conditions,researchers use the rhesus monkey facial action coding system Maq FACS to mark the facial expressions of rhesus monkeys in an objective and standard way.However,at present,most of the facial expressions are manually marked,which is time-consuming and laborious and has certain errors.With the continuous development of machine vision image processing technology,automation has become the future development trend.In this paper,we propose a method to automatically mark the facial expressions of rhesus monkeys using facial action coding system.The main research work and contents are as follows:(1)Aiming at the problem of monkey face region detection,a set of monkey face region detection algorithm suitable for rhesus monkey face structure and effective removal of interference information was established.The interference information included the monkey’s facial hair and the facial displacement caused by the monkey’s head movement,rather than the subjective movement of the monkey’s five senses during the expression.A method combining the Retinaface network with affine transform is proposed to detect monkey face.According to the facial structure of rhesus monkey,the target frame of monkey face was adjusted according to pupil distance ratio of monkey,which effectively removed the interference information and obtained the data set of monkey face facial expression feature extraction.(2)For feature extraction,this paper proposes the method of combining geometric features with texture features.The geometric features are composed of spatio-temporal features and vector features,and the spatio-temporal features are constructed by the displacement of 19 key points on the monkey face between frames in the continuous video to describe the movement trend of the monkey face.Vector features are constructed by combining 19 key points in the monkey face in the video frame to describe the muscle contraction of the monkey face.Texture features are composed of Gabor wavelet transform to perform eight directions and five scales of filtering transformation on ROI of key points to describe local texture changes of monkey face.Adaboost algorithm was used to fuse geometric features and texture features to obtain more comprehensive feature information for recognition of monkey face expression.(3)Four expressions of rhesus monkeys were coded through the facial action coding system of rhesus monkeys.Eight facial motion unit classifiers were trained using geometric features,texture features and fusion features,and the classification results were compared.Finally,the facial motion unit classifier trained by fusion features is used to classify and recognize the four expressions of monkeys.In this paper,good recognition effects are achieved for the strong and weak expressions of monkeys.