Face Recognition Based On Convolutional Neural Networks

As a research hotspot in the field of computer vision,face recognition has played an increasingly important role in real life in recent years.Face recognition under natural scenes and its application based on Convolutional Neural Network(CNN)are mainly studied in this paper.Compared with traditional face recognition methods,the method based on CNN has great advantages.First,there is no need to extract facial features manually from face images and the facial features will be automatically extracted from the picture by convolution operation,during which local perception and weight sharing play an important role.Then,if the constructed network model is reasonable,the trained network model will have a good recognition effect after repeated training and learning through the samples.Since that the structure of the network has a great influence on the recognition effect,this paper focuses on the rationality of the network model to make the model can converge quickly and stably on the sample set through simulation experiments,thus obtaining higher recognition accuracy and good performance in practical application scenarios.The main contents of this paper are:1)The development history of CNN and its application in face recognition are introduced.At first,the traditional methods of face recognition are described;Then,the basic theoretical knowledge of face recognition using CNN is introduced,and the compositional structure and basic characteristics of CNN,as well as the various calculation layers of CNN are elaborated;Finally,the application of several CNN application cases is introduced to illustrate the basic structure of convolutional neural network model and its application scenarios.2)A convolution neural network model for face recognition is constructed.The AR face database and ORL face database are used to train the model and the test results show that the network structure can achieve higher recognition rate.And the effects of activation parameters,number of convolution kernels,dropout,momentum term and other parameters on network performance are studied through experimental simulation.On this basis,the original network model is improved to make the performance and recognition effect of the model further better,and the simulation results show that the recognition rate of the improved model for AR face database and ORL face database reaches 99.78% and 99.82% respectively.3)Using the network model constructed above,a browser-based face recognition system is designed in this paper.The system can create a face database by uploading pictures or by capturing pictures using camera and can quickly accomplish the task of face recognition with extremely low hardware requirements.Experimental results show that this system has a good recognition effect on normal human faces and can perform some recognition tasks on human faces with insignificant features.