Expression is an important embodiment of human inner emotion change.Current expression recognition methods usually rely on global facial features,ignoring local features extraction.Psychologists point out that different facial expressions correspond to different regions of local muscle movement.In this paper,we propose an expression recognition algorithm based on local representation,referred to as EAU-CNN.In order to extract the local features of the face,the whole face image is first divided into 43 sub-regions according to the 68 feature points of the face.Then,8 local candidate regions covered by the muscle motion region and the facial salient organs are selected as the input of the convolution neural network.In order to balance the features of local candidate regions,the EAU-CNN adopts 8 parallel feature extraction branches,each of which dominates the full connected layer of different dimensions.The outputs of the branches are adaptively connected in terms of attention to highlight the importance of different local candidate regions.Finally,the expressions are divided into 7 categories:neutral,angry,disgusted,surprised,happy,sad and afraid by the Softmax function.In this paper,the algorithm is evaluated on CK +,JAFFE and custom FED datasets.The average accuracy of the proposed algorithm is 99.85%,96.61% and 98.6%,respectively.The evaluation index is 6.01%,10.17%,6.09% higher than that of the S-Patches algorithm.The results show that local representation can improve the performance of emotional recognition.
