Research And Application Of Face Recognition Technology Based On Deep Learning

The development of the society has endowed the identity with great significance.Facial information which everyone has features nature and advantage that it can't be perceived by detected individual.Consequently,face recognition has being one of the research focus in computer vision.And it has been widely applied in the fields of security,finance,education,e-Government affairs.Deep learning has shown great advantages in the research of face recognition technology,especially convolution neural networks.Nevertheless face recognition is influenced by illumination,occlusion,facial expression,aging and such variation.So there is still more room to improve it in the aspect of technology.In this thesis,we are directed against the special characteristics of comparison of identity card and person,designing the face recognition algorithm for the application,aiming at the accuracy and performance of the algorithm.The main contents of this thesis are as follows:(1)First,related knowledge of face recognition and deep learning are introduced.And then,convolutional neural network structure is illustrated specially,including the distinctive design of the network such as sparse connectivity and parameter sharing.The function of pool layer and convolution layer is also showed in this thesis.Then the typical convolution neural network models are analyzed and compared.(2)In this thesis,each part of the face recognition process is carefully designed and analyzed.Our application of face recognition is influenced by illumination,so we do experimental comparison and analysis about the methods of face detection and image preprocessing.By taking targeted approach,we got a strong robustness model with the illumination fluctuations.(3)ResNet is one of the typical convolution neural network.By adjusting the depth and structure of ResNet and combining with different loss functions,a new network model structure is designed in this thesis.And the model has showed a better performance in terms of the accuracy,size of the model,and time of training.(4)In order to get a better model for face recognition,we built an Asian face datasets and an Asian face test set by ourselves.And we got a better model based on the network and data.Experiments showed that the model we trained on our datasets do well in the test set and real application.(5)Finally,a functional comparison system is designed and implemented based on the above algorithm model.This thesis gives a detailed introduction and description of each function module in the system and the important process.The system verifies the advantages of our algorithm and meet the requirements of real application.