Based On Global And Local Face Occlusion Recognition Algorithm

Nowadays,the development of the face recognition neighborhood is faster than ever,many scenes in life have applied face recognition technology,such as face payment,face access control and so on.The development of face recognition technology has brought a lot of convenience to people,but today's face recognition still has many deficiencies,these deficiencies also restrict the development of face recognition.The current face recognition must ensure that the face is not blocked by an object during the process of recognizing the face.If there is an object,the face cannot be recognized.This problem has caused a lot of inconvenience to people,and it is urgent to develop to solve the problem that the face can be correctly recognized when it is blocked by an object.In order to improve the problem of low recognition rate in the case of occlusion,this paper proposes a global and local occlusion recognition algorithm that uses deep learning for feature learning.Therefore,the main research results of this paper have the following parts:1.This paper first proposes the concepts of global and local faces.Traditional occlusion recognition methods only focus on the recognition of global faces while ignoring local factors.Due to occlusion,local prominent features will be missing,and the impact will cause global the accuracy of recognition decreases,so it is difficult to improve the accuracy of recognition.In order to improve the accuracy,this paper will consider the global and local factors comprehensively,the training set uses the AR data set.In this paper,SIFT and SVM algorithms are used to discriminate the parts of the face on the data set,and the blocked partial faces are cut off to form a global face.The partial faces are mainly facial features.Then the global and local data sets will form the final training set.2.In this paper,based on the concepts of global and local,an algorithm of global and local face occlusion recognition mechanism based on attention mechanism is proposed.First of all,this paper will use the lenet-5 convolutional neural network.During the training process,different weights will be given to the global and local.For example,if the eye is blocked,the weight of this part will be reduced.This training greatly strengthens the other.Occlusion of local details has greatly improved the accuracy of training.3.In this paper,a multi-classifier similarity weighted voting method for face occlusion recognition is proposed based on the concepts of global and local.This article will create multiple classifiers for global and local data sets for training.The final classifier will use the improved layered voting method to make a final decision on the face,speeding up the training of multiple classifier and forecast speed.