Research On Face Image Recognition Based On Convolutional Neural Network

Biometrics plays an important role in our lives.Face recognition technology is a very hot research field in biometrics.With the development of deep learning,the face recognition technology based on Convolutional Neural Network has become the main method of face recognition.Based on this,this paper based on CNN's face recognition technology research and implementation.The main contents are:The basic principles and implementation of CNN is introduced.By studying the basic principles of the CNN model,using the convolution function and the downsampling function contained in the Matconvnet framework to process the image,the results obtained are used to construct the convolutional and downsampling layers of the convolutional neural network model,thereby building the required CNN training model.In addition,the understanding and mastery of the full-connection layer and the classification layer were introduced through in-depth exploration of the basic principles of the multilayer perceptron MLP.All of the model's implementation uses the MATLAB-based Matconvnet framework,along with its library files,and GPU acceleration.This paper builds,analyzes and trains the face recognition model based on CNN.An improved CNN face recognition network structure diagram is designed and researched.The original design network consists of 10 layers,including 5convolutional layers,4 pooled layers,and 1 fully connected layer.The network model is extended to the coarse model.Based on the foundation,a group of convolutional layers and pooled layers are added to form a new fine network.In the course of meticulous training,the ninth layer of the original network is connected with the newly added convolution layer to extract deeper features and design training.The internet.Two models of thickness were used to train on the face data set,and fine tuning parameters were used to optimize the model parameters.Based on the trained network,facial recognition training and testing were performed using images in the face image database.For ordinary face images,coarse network recognition was over90%,fine network was up to 93.8%,and it was superior to the algorithm.Forcross-age face images,the precision of the fine model test reached 90.2%.In addition,this paper designs a face recognition framework that includes face location and segmentation algorithms,and describes the process of image preprocessing and CNN-based face recognition in detail.Tested on an image dataset containing the human torso and background.The recognition rate reaches more than91.7%.