The Research And Application Of Face Recognition Based On Convolutional Neural Network

With the development of artificial intelligence,how to identify users accurately and effectively and improve information security becomes an important research direction.Compared with traditional rely on items and their own memory(such as card and password)certification recognition methods possible hazards and the insufficiency,then put forward,such as fingerprint recognition,iris recognition,DNA recognition and face recognition based on the identification method of biological technology.Compared with other biometric methods,face recognition has the advantages of non-contact,non-mandatory and concurrency,which is more acceptable to users.n recent years,in areas such as education,e-commerce,scholars have put forward a lot of face recognition algorithm based on machine learning and pattern recognition,the recognition accuracy of some of the algorithm has reached a quite high level,but in practical application,it will be subjected interference and restriction by some kinds of external unavoidable factor sometimes.This article first introduces the SVM based on deep learning combined with facial recognition technology,with the aid of matter unsupervised automatic feature extraction ability to deal with image data,then introduce the generalization ability and kernel function to deal with nonlinear classification ability of support vector machine(SVM)model to simulate the face image classification.In the CMU-PIE face database,the recognition accuracy was increased to 91.2% of RBM-SVM model by 80.7 and 87.3% of the SVM model.The influence of external factors on recognition accuracy was analyzed by experiment.In the real environment,face images were collected by equipment,has a strong variability and complexity of these images are highly susceptible to collect Angle,posture,and the influence of illumination conditions,these are has brought great negative impact for face recognition accuracy.For such problem,this paper then introduces the has certain translation and rotation invariance of convolution neural network(lenet-5)face recognition methods,and on the Caffe platform using the GPU to CMU-PIE experiment of 41386 face image data in the database,get a facial recognition accuracy can reach 96.1%.On this basis,the paper then puts forward a new,improved parallel synchronous convolution neural network model(CNN-LP2),input the original images and enhance light double into CNN-LP2 model,get the recognition accuracy as high as 99.2%.In the improvement of the various parameters of the model,this paper further guarantee model at the same time,more shows the proposed CNN-LP2 model in dealing with the superiority of the face image recognition.