Research On Real-time Recognition Technology Of Human Facial Expression Based On Image

In daily life,the simplest and most effective way of communication between people is face-to-face communication,but in addition to language communication,facial expressions can also convey and express individual information to achieve the purpose of communication.Nowadays,in order to pursue a better human-computer interaction experience,expression as an important means of communication has gradually become a key research topic.Although traditional facial expression recognition algorithms have good performance in small sample tasks,facial expressions are complex,and the differences between individuals cannot be described by the underlying features alone,resulting in low recognition accuracy and poor generalization ability.In addition,traditional recognition methods require researchers to manually extract features,which requires a lot of experience and time to adjust the relevant parameters of the model and consumes a lot of manpower.With the continuous development of deep learning,the end-to-end model structure is designed to avoid the influence of human factors.Through the combined action of the convolutional layer,activation function,and pooling layer,the expression information is gradually abstracted,and the resulting features have strong Discriminative ability.Therefore,in order to improve the accuracy and generalization of expression recognition,this paper proposes an expression recognition system based on the fusion of global multi-scale CNN features and local Landmark-SIFT features.The specific work is as follows:1.Design a custom multi-scale convolutional neural network,use different scale convolution kernels,integrate multiple scale information to capture the potential relevance in the image,and improve the model's discrimination and generalization ability.In order to reduce the model parameters and reduce the amount of calculation,the batch standard speech layer(BN layer)and the Dropout strategy are introduced.At the same time,the central loss function is used to reduce the difference between classes and improve the generalization ability of the model.2.Because the existing facial expression database samples are few,the traditional manual feature extraction algorithm has better performance for small samples.Therefore,this paper designs an expression recognition algorithm based on the fusion of global multi-scale CNN features and local Landmark-SIFT features.The SIFT descriptor has a good rotation scale and invariance to part of the illumination.After the face image is calibrated by the facial structure points,the SIFT descriptor can be extracted to concentrate the key points on the eyes,eyebrows,and mouth and other areas that contribute a lot to expressions.A visual bag-of-words model is constructed to standardize the local features,and finally the obtained features are fused.3.Design a real-time facial expression recognition system.This system can recognize static pictures,dynamic videos,and real-time images collected by cameras.The system can provide real-time feedback based on facial expressions.