Research And Application Of Deep Learining In Face Recognition

With the big data development of human society in the 21st century,deep learning as the representative of machine learning,has gained wide researchers' attention,showing its huge practical value.Deep learning is widely used in daily life with its kind of brain and abstract characteristics.Face recognition is one of the widely used fields in deep learning.In a large number of deep learning algorithms,convolutional neural network has achieved remarkable results.Other deep learning methods such as support vector machine have also promoted the development of face recognition technology and have gained popularity from researchers with its superior performance.The thesis focuses on two aspects:feature extraction and classifier optimization.The purpose of this thesis is to improve the ability of fusing global and local features and combining multi-layer features of deep learning algorithm and provide stronger nonlinear classification ability for linear classification algorithm.These can help improve the stability and recognition rate of algorithm.First,in chapter one,the research background,development and application of deep learning are introduced.Then,it is introduced that the problems and challenges of face recognition at this stage.At the end of this chapter,the content arrangement of this thesis is briefly described.Then,in the second chapter,a face recognition algorithm based on convolutional neural network for classifier optimization is proposed.Because of the linear classifier between the fully connected layer and the output layer of convolutional neural network,it is difficult to achieve good classification results.The support vector machine is a good classifier,and the artificial bee colony algorithm can avoid falling into local optimal location.Therefore,in this chapter,the support vector machine optimized by artificial bee colony algorithm is used to replace the linear classifier of convolutional neural network.This approach can improve the accuracy of the classification.In the third chapter,a face recognition system based on fusion of global and local features is proposed.The system is intended to solve the deficiency that the image features extracted by single method are difficult to express the facial information comprehensively.The global and local features of images are extracted respectively by using different approaches.Finally,the classifier decision-level fusion is implemented to get a more comprehensive expression of facial features.The system not only achieves better classification results,but also improves the stability of the algorithm.In the fourth chapter,a face recognition model based on multi-layer convolutional neural network features and classifier optimization is proposed.In consideration of the fully connected layer features of convolutional neural network are difficult to express the facial information completely and the structure of convolutional neural network also lacks a proper classifier.The innovative approach proposed in this chapter can be described that multi-layer convolutional neural network features are extracted to enhance the image representation.The optimal classifier is used to improve the face recognition accuracy and efficiency of the model.Finally,the fifth chapter of the thesis makes a summary of the work and prospects for the future research.