Research And Implementation Of Face Recognition Method For Partial Occlusion

With the rapid development of the Internet and the landing of artificial intelligence technology,face recognition technology has been widely used as an identification technology in various fields.With the development of deep learning theory in recent years,many new models have emerged.The algorithm constantly refreshes the accuracy of face recognition.However,these high recognition rates are often based on specific conditions.Under non-limiting conditions,especially when there are strong light conditions and occlusion,the accuracy of face recognition decreases sharply.For the occlusion situation,scholars have proposed and demonstrated many methods.Although these methods have proved to be effective,they still fail to achieve very good recognition rates.These methods almost all use traditional manual feature design and classifiers.In the era of deep learning,there is still much room for development in face recognition based on occlusion.This paper mainly focuses on the situation that the lower part of the face is occluded.This paper proposes an algorithm based on convolutional neural network and regional weights.In order to compare with the algorithm proposed in this paper,this paper first introduces the principle of the traditional subspace regression method,introduces and tests two sub-algorithms,and then introduces the basic principles and structure of the convolutional neural network,and introduces the VGG16 network model and VGG FACE model,and then analyzes the receptive field range of the Feature Map output from the convolutional layers at each layer of the VGG16 network.According to the receptive field theory and basing on the idea of regional weights,a modified truncated model is proposed.After this model is applied to Feature Map,gives different regions with different weights,which can effectively reduce the influence of occlusion features and retaining the features of the none-occluded face region as much as possible.Experiments prove that the proposed algorithm is effective.The experiment 1:1 face authentication and 1:N face recognition are better than the original VGG FACE model,and the effect is much better than the traditional method.Based on the algorithm proposed in this paper,we designed and implemented a face recognition attendance system,including personnel information management,attendance information management,attendance information record.In the development process,the requirements of the attendance system are first analyzed,then the logical architecture and functional architecture of the system are designed.Then the database ER diagram and database table are designed according to the actual functions,and then the tools such as Caffe+Pyqt+Qt Designer+Mysql are used.Finally,the test case was designed.The function and performance of the system were tested on a self-built face database CAE-FACES.The test results showed that the system has strong accuracy and practicality.