Research On Occlusion Face Recognition Algorithm Based On Nuclear Norm Robust Regression

The face recognition technology in controlled environments is relatively mature,but the face recognition problems in complex real environments still face many challenges that are difficult to solve.The main reason is that the structure of face images in real scenes is unstable,these face images are often affected by occlusion,lighting,facial expressions,human posture,and other factors,which makes the task of face recognition more difficult.The method based on nuclear norm regression mechanism are widely known by scholars.However,these methods focus on the low-rank structure information of error images,ignoring the correlation of sample image representation,which leads to the decline of recognition performance when dealing with face recognition problems affected by multiple complex factors.Therefore,this article proposes two improved algorithms to solve such problems.The main work is as follows:(1)To solve the problem that the existing algorithms based on nuclear norm regression mechanism degrade the performance of occluded face recognition in complex environments,this paper proposes the matrix regression model with relaxed block diagonal representation.From the perspective of obtaining discriminant representation,the model introduces relaxed block diagonal representation into nuclear norm matrix regression,which can dynamically optimize the block diagonal component of the representation matrix to strengthen the intra-class correlation and inter-class incoherence of sample representation.Furthermore,the training samples and testing samples representation is formulated as a joint optimization problem to strengthen the coherence of sample representation.At the same time,nuclear norm constraint is used on error image to capture low-rank structural information about the error image,reducing the interference of that portion of face recognition.The experimental results on Extended Yale B,LFW,AR face data sets show that the proposed algorithm has good recognition performance for occluded face recognition in complex environments.This advantage is particularly evident when the continuous occlusion ratio is large.(2)In complex environments,the error distribution is more various,and face images often face multiple influences that increase recognition difficulty.This paper proposes a robust matrix regression model with local constraints.First,nuclear norm and Frobenius norm constraints are used to obtain error image structural information.By combining these two items,it achieves the robustness of occluded face recognition under the influence of complex noise.In addition,from the perspective of obtaining more discriminative representations,the proposed algorithm make full use of sample label information and direct constrain the representation components,which are closer related to the recognition process rather than the representation coefficients.Futhermore,the locality structure of characterized by subspace distance is utilized to learn the weight of a class,which can enhance intra class correlation and obtain a more discriminative representation for classification.Various types of experiments simulated on Extended Yale B,ORL,and AR face datasets show that the proposed algorithm can effectively solve the problem of occluded face recognition in complex environments and has the robustness to occlusion.