Facial Expression Recognition Of Image Sequence By Graph Convolution Neural Network

In daily life,facial expression is an indispensable carrier of information when people communicate with each other.In the information received by the listener,facial expression more than half of it.In recent years,with the intelligent life and the expansion of robot market,service robots gradually replace traditional artificial labor force.Facial expression recognition is an indispensable function to improve the interaction ability of service robots and make robots can provide high-quality services like people.Therefore,face expression recognition algorithm has gradually become a research hotspot in computer field.Facial expression recognition algorithm can be divided into facial expression recognition based on single image and expression recognition based on image sequence according to the processing data.Expression is the process of facial muscle change.In daily life,people usually record the change of face through monitor and get the video of expression transformation.Therefore,the expression recognition algorithm based on image sequence is widely used in life.This thesis proposes a new algorithm based on graph convolution neural network to recognize the expression of image sequences.This algorithm can be divided into three steps.Firstly,the expression intensity evaluator is trained by Siamese neural network,the expression intensity is carried out for each frame image in video,the similar intensity image is removed,the new image sequence is extracted to reduce the calculation cost.Then we construct and train the convolution neural network of the attention mechanism.We use the convolution layer after training to extract the spatial features of each image in image sequence to form feature matrix.Finally,the feature matrix is used as the input of the graph convolutional neural network to recognize the image sequence.The algorithm is verified on CK + data set,and compared with other algorithms.The results show that it has high accuracy and excellent recognition effect.