A Study on Multi-pose Face Recognition Based on Improved Local Convolutional Neural Networks

In view of the problem of attitude change affects face recognition rate, a modified local convolutional neural network is proposed. Firstly, a new convolutional network based on Resnet50 is built to maximize the extraction of image features, and then the data transmission rate is used to reduce, and the number of sampling points to prevent the distortion caused by aliasing. Finally, the method of mapping posture correction is used to obtain the local texture information of the positive face picture obtained in the top-level reconstruction. The experimental results show that the algorithm has significantly improved the recognition accuracy by 1.13% on the face data set LFW compared with the improvement, which makes the face frontal recognition more flexible and accurate.