An Improved Method Based On Sparse Representation For Robust Face Recognition With Occlusion

Face recognition is a very important and valuable technology in computer vision community and image processing field.This technology mainly aim to identity authentication and identity information retrieval based on face image information.Due to the efficiency and intelligence of this technology,numerous face recognition methods have been proposed and widely used in financial payment,enterprise punch card,security and other fields.In the past few decades,due to the continuous research and improvement of theoretical knowledge related to artificial intelligence by many scholars and the continuous practice verification of face recognition technology in the industry.It makes the face recognition technology more and more mature and the accuracy is higher and higher.However,there are also some challenges remain.For example,when the face is blocked,especially in large areas,the recognition accuracy will be significantly reduced.In this paper,an improved face recognition method based on sparse representation model and Gaussian mixture method is proposed to solve the problem of low recognition rate in existing face recognition systems,especially in traditional face recognition methods.First of all,the training set of sample in the same position pixel information of the Gaussian mixture model and through the expectation maximization algorithm to estimate the parameters of Gaussian mixture model.Secondly,the weighted European distance and threshold operations are combined to determine face occlusion to achieve the parameterization of face occlusion information,so as to more accurately and conveniently retrieve the missing part of face information caused by the interference of face occlusion.Verification experiments on AR data set show that this method is sensitive to the information of the occlusion interference part of the face,and has improved the accuracy and robustness of occlusion detection compared with traditional occlusion detection methods(such as edge detection and image segmentation).Finally,in the process of face recognition and classification,the influence of the occlusion information of the samples is considered The occlusion information judged by the face and the undisturbed information are separated,and the face after the occlusion information is removed is recognized by sparse representation classifier,so as to realize the accurate recognition of the face image with occlusion.Gaussian mixture model applied to the face and fine-grained to each pixel to more accurately detect the human face due to cover the missing part,at the same time willkeep out after the interference caused by information extracting sparse representation and classification,as far as possible to avoid the shade the compensation due to the face or artificial subjective error introduced by the results of face recognition classification of interference,this will keep out information for parameterized mentality to face block identification follow-up research opens up a new train of thought,at the same time to improve the accuracy of face block identification and robustness has very important practical significance.