Research On Masking Face Recognition Algorithm Based On Deep Neural Network

The outbreak and spread of covid-19 has severely impacted the daily lives of people around the world.To prevent the spread of COVID-19,the World Health Organization recommends that people wear masks in public places.Wearing a mask will shield the important feature information of the face,resulting in the loss of important features of the face,which poses a severe challenge to the traditional face recognition technology.Therefore,it is particularly important to use algorithms with simple implementation process and high recognition efficiency to identify faces obscured by masks.This paper combines deep neural network and attention mechanism algorithm to achieve effective recognition of masks covering faces.The main research work is as follows:(1)For the public face dataset,there are problems such as unbalanced data and insufficient number of masks.In this paper,multi-task Cascaded Convolutional Networks(MTCNN)are used for face detection,removing redundant sample information and reducing network computation.The Dlib algorithm is used to delineate the occlusion space for human face features,supplemented by affine transformation method,which maps the occlusion mask module to the predetermined position of the face,and generates the masking data set Maskface.Study the face recognition method of different deep neural networks and conduct comparative experiments.The experimental results show that the deep network structure can obtain enough discriminant features to improve the recognition accuracy,but it is still impossible to solve the mask masking problem.(2)For the problem that the accuracy of face recognition under the mask is not high.this paper proposes an improved algorithm based on deep residual network(ResNet),which introduces convolutional kernel replacement thinking and average pooling layer in the network structure to reduce network parameters and alleviate information loss;secondly,the Mish activation function is employed to increase the network’s generalization capabilities;and lastly,the joint loss function is introduced,which makes the face class spacing more distant and the class spacing more compact to improve the face recognition accuracy.Experimental results show that the proposed algorithm is effective for masking face recognition problems.(3)In order to further improve the accuracy of masking face recognition,In this paper,a masking face recognition algorithm based on attention mechanism is proposed,which embeds the attention mechanism network in the ResNet network structure,gives higher weight to the non-masked area,and gives lower weight to the masked features,so that deep feature extraction of the face may be achieved in order to increase the training effect experiment findings reveal that the suggested technique enhances the accuracy of masking recognition significantly,and ensures the versatility of general face recognition.