Occlusion Face Recognition Based On SR-NMR And GD-HASLR Algorithms

In the field of pattern recognition,face recognition methods have always been a very popular research direction,and excellent research results are emerging one after another.In recent years,with the rapid development of technology and people’s increasing demand for identity verification,face recognition technology has been widely used.However,there are still some problems in the application of face recognition technology to real life,because many current algorithms are basically carried out in an ideal experimental environment.When there are some occlusions on the face,such as scarves,glasses,sunglasses,posture,and masks,the performance of many algorithms will drop sharply.Especially in recent years,the new crown epidemic has spread all over the world,and wearing masks has become a necessity for people’s social life.Therefore,this paper proposes an occluded face recognition method combining sparse regularized NMR and collaborative representation.In addition,most of the existing algorithms are carried out with sufficient training samples,while face recognition in real life generally only selects a small number of training samples.Therefore,when the number of training samples decreases rapidly,the accuracy of many algorithms also drops sharply.This paper proposes an improved GD-HASLR method.The specific contents of this paper are as follows:（1）Aiming at the problem of occlusion of faces in images,this paper proposes an occluded face recognition method that combines sparse regularization NMR and collaborative representation.First,the sparse regularization NMR algorithm is used to obtain the sparse representation of training samples and test samples.the low-rank structural information and discriminative representation of the error image can be obtained from the sparse regularized NMR model,and the sparse and distinguishable representation can be obtained by using the L₁ norm,and the sparse representation obtained after removing the occlusion in the image is more separable,and finally use the collaborative representation classifier to classify the sparse representation coefficients of the test samples.The experimental results show the effectiveness of the method.（2）In view of the small number of training samples,this paper proposes an improved GD-HASLR method,which first calculates the gradient size and gradient direction of the face image from the first order to the third order,and then uses the sigmoid function to map to obtain the gradient.The direction vector is used as the input of the hierarchical sparse low-rank model,and the sparse representation coefficient and error vector are optimized and solved.Finally,the residuals of the first-order to third-order test samples are calculated respectively,and the category with the highest frequency or the lowest average level is selected as the identification result.The experimental results show that the improved GD-HASLR method in this paper has higher accuracy than the original algorithm.