Research And Application Of Deep Learning In Face Recognition

Face recognition has extensive application and potential development space in the field of security,financial,e-government affairs,entertainment and so on.At the same time face recognition technology based on large data become a hot topic with the development of deep learning and lead to a lot of work on numerous scholars.This paper mainly researches two important algorithms of deep learning,and proposes an improved algorithm to solve the problem that the recognition rate of deep learning model in small sample database is low.This paper set up a small sample of face database and gives full consideration to the samples of the change of the expression,posture,illumination and so on.By switching the face data from the RGB color space to YCbCr space,we can get the face data after scale normalization.Analyzing the basic principle,model structure and generation algorithm of CNNs and DBNs and establishing respectively the face recognition models.After making experiments on Extended Yale-B?ORL and self-built database and then analyzing their advantages and disadvantages,we can draw relevant conclusions.For the lower precision of small database than large database in both CNNs and DBNs,and deep learning model for big data,an improved DBNs model has been proposed.First using local binary pattern processing data,then get the LBP histogram as the input layer of DBNs and introduce the pairwise classification model in the stage of parameter optimization using BP networks.It can optimize the system classification boundaries,and improve the recognition effect.The experimental results in three databases show that the improved face recognition algorithm also has a good recognition effect on small sample database.When the data have all kinds of interference,the system can also be well to classify.Finally designing a simple GUI interface to realize the application of face recognition.In this paper,the recognition rate of the improved algorithm in the standard database is 100%,in the self-built database is 96.5%.