Research And Implementation Of Face Recognition Based On Deep Learning

As a frontier topic in the field of biometric recognition,face recognition technology has been the focus of academic and industrial research.In the past decades,researchers have proposed many different face recognition algorithms.In recent years,deep learning is hot,many face recognition methods based on deep learning show strong performance.This paper studies the face recognition method based on deep learning at the present stage,and has done some exploration on the deep hash,and completed the following work:(1)The development history and research status of deep learning,deep face recognition and deep hash are summarized;the structure and principle of convolutional neural network are introduced;the source code and software architecture of Caffe are analyzed,which is a open source deep learning framework,and how to use Caffe to carry out deep network training and image recognition tasks are introduced combined with MNIST handwritten digit recognition.(2)Aiming at the problem of large number of sample data requirement and insufficient training data,based on the idea of transfer learning,this paper use the small sample dataset to adjust the existing deep face model,and transfer it to specific face recognition tasks.On the LFW face verification set,our method based on transfer learning has obtained 97.52%precision,which shows the effectiveness of this method.(3)Aiming at the problem of high dimension and huge computation cost of existing face recognition methods,this paper introduce the deep hash to transform the image into a class of perceptual hash codes to achieve face image retrieval,recognition and matching.In order to reduce the clash of the image hash index and improve the accuracy of the face image retrieval,a method based on deep hash and metric learning is proposed.This method uses deep neural network to learn the deep features of images,and introduce the thriplet loss function,so that the intra-class distance is as small as possible,and the inter-class distance is as large as possible.Then,we use random mapping to map the high-dimensional face features in the feature space to Hamming space,so that the distribution in the Hamming space can be dispersed as much as possible,so as to reduce the collision probability when querying.The experimental results show that our method has better performance compared with the existing methods.(4)In view of the application of the gate check,the face recognition and checking system is developed,and the development process and software architecture are introduced.At the same time,an auxiliary module is developed for experiment in the real scene,and the feasibility and effectiveness of the system are verified.In addition,in order to facilitate the transplantation,the main functional modules are encapsulated,and the packaging interfaces and methods are introduced.Finally,this paper give a summary,and look forward to the future research direction.