Research On The Application Of Face Recognition And Deep Learning

As an important means of personal identification,biometric technology is emerging all the time in the daily life of human beings.Compared with other biometrics,face recognition is widely applied in the field of information security because of its simplicity,intuition and shortcoming.The privacy of facial features has also become the focus of public attention.How to scramble the face image and recognize it directly in the scramble domain has become a hot research topic,and the impact of insufficient face feature extraction on the recognition rate is also a challenging research topic.Traditional shallow neural network has been unable to accurately express the relationship between facial features in complex and changeable situations.Meanwhile,the inadequate extraction of sample features can also cause the failure of network training.In this paper,a new face recognition system based on the convolutional neural network(CNN)of deep learning is designed,which aims at face privacy protection requirements.First,a scrambling operation is implemented on the human faces,which caused that the human face can not be observed in the image.After that,face recognition is applied by the convolutional neural network in the scrambling domain.To improve the feature extraction,the feature fusion method is utilized to strengthen the features,and a hybrid face recognition network model is designed to recognize the enhanced image.In this model,the convolution neural network is used to extract the features of the image.Finally,the restricted Boltzmann machine is performed to classify the samples.Thanks to the excellent recognition effect,a multi-environment wireless authentication system based on face recognition show a strong security.The specific work of this article is as follows:(1)CNN face recognition based on the facial privacy protection.For human face privacy protection,Arnold transform is used to scramble face images,which is a recoverable periodic transformation.Then,four-layer convolution neural network face recognition model is designed.To improve the recognition rate of the system,the Arnold transform of the face image is advanced by taking random values of the number of different faces in the common face library.Experiments validate that the algorithm and network model proposed in this paper have good experimental results in ORL,PIE and self-mining face library.(2)Feature fusion face recognition based on hybrid neural network.To accommodate the insufficiency of feature extraction,a faster R-CNN is applied first to detect faces and human ears in the picture.Then,the wavelet image fusion method is used to fuse the detected images and form a new feature enhancing training set.After that,the feature extraction and dimensionality reduction of the convolution neural network are performed for feature extraction.Finally,the restricted Boltzmann machine is used for classification and recognition.The experiments shows that the algorithm and network model proposed in this paper have good recognition effect.(3)The design of multi-server authentication scheme based on face recognition.Combining the two algorithms and network models,a multi-server authentication scheme is designed based on chaos mapping and facial features,which can significantly improve the safety and anti-cipher guessing of multi-server authentication system.