Research On Face Recognition Based On LBP Deep Learning

As one of the biometric identification technologies,face recognition technology has become a research hotspot in the field of pattern recognition.Aiming at the problem of the insufficient feature extraction and low accuracy of face image recognition,a new method of image recognition is developed by the improvement of local binary pattern and combining with the deep learning model.The main results of this paper are as follows:1)Feature extraction of face image based on improved LBP algorithm.Aiming at the problem that the traditional local binary pattern can not extract more effective features,this paper proposes a new algorithm.The sampling range of the original 8 neighborhood is extended to the sampling range of 16 neighborhood,which can extract the more expressive features and avoid redundant information.2)The method of preventing over-fitting based on Dropout.Aiming at the problem that deep learning is a kind of machine learning model with deep structure leads to over-fitting problem easily,therefore the dropout technology is added to train DBN,freezing the parameters of some neurons whose output is zero with a certain probability at each training time to avoid an over-fitting feature and improve the generalization ability of the network.3)Face recognition method based on LBP and deep belief network.Aiming at the problem that DBN can not make full use of local features of human face,the improved LBP feature is used as the input of deep belief network with dropout.Finally,the recognition of face images are performed by trained networks.The feasibility of the proposed algorithm is verified by experiments,and the high recognition rate is achieved.Improved algorithm can effectively extract the image texture features and it has a good recognition effect in face images,which opens up a new idea for the research of face recognition.