Research And Application Of Facial Expression Recognition Algorithm Based On Tensor Representation And Deep Learning

In the present highly intelligent modern society,more and more fields begin to take biological identification technology as an important means to meet the needs of the current society and life.Facial expression recognition,as a kind of biological recognition technology,enables machines to perceive and analyze human's inner complex emotions just like human beings,which greatly promotes the development of biological recognition technology towards strong artificial intelligence.However,due to the inherent complex patterns of facial expressions,many existing methods for biological recognition are unable to distinguish various expressions effectively.Therefore,this paper focuses on the tensor representation and deep learning algorithm of facial expression recognition to improve the accuracy of facial expression recognition.The main work of this paper is summarized as follows:1.This paper introduces the facial expression recognition based on tensor representation,and systematically expounds the relevant tensors theory,the dimensionality reduction algorithm and the classification algorithm needed in the facial expression recognition process..2.The expression recognition algorithm based on low rank representation and tensor decomposition is studied.In order to solve the problem that the samples in the data space can not be effectively classified due to the influence of identity and are easy to be affected by noise,this paper introduces a low rank representation model and establishes a low rank reconstruction algorithm based on tensor,so as to map the discrete tensor expression space to a more discriminative reconstruction subspace.However,in the process of constructing tensor data,the overall dimension of the data increases with the number of orders,resulting in the problem that the existing classification methods cannot be adapted.The dimensionality reduction algorithm is considered to process tensor data.Nevertheless,traditional dimensionality reduction method is easy to destroy the internal structure of data,so this paper studies the expression feature extraction method based on tensor decomposition,using the non negative characteristics to effectively realize the representation of the whole from the base image structure Tensor data.3.The algorithm of facial expression recognition based on deep stacking network is studied.Firstly,the stacking results are built on the basis of depth neural network,and the hidden features in the network are learned by ridge regression.In order to solve the problem that the traditional deep learning needs a lot of data and massive training,the feedforward neural network is used to simply realize feature extraction and classification without back propagation.Experimental results show that the simple method in this paper is robust,which improves the recognition accuracy of facial expression image.