Low Resolution and Multi-pose Face Recognition based on Residual Network

The purpose of low resolution face recognition (LRFR) is to recognize faces from small or low-quality images with different poses, illumination and expressions. However, the recognition effect of LRFR is not ideal under uncontrolled conditions. In this paper, residual network of resnet18 is used to test the recognition rate of six groups of UMIST face datasets with the resolution from 112×92 to 3×3. The results of recognition rate under different resolution states are compared and analyzed, and the dynamic curve of recognition rate with the change of resolution is obtained. In order to solve the problem that the recognition performance declines sharply when the resolution is lower than the minimum resolution, The minimum resolution image is reconstructed by super-resolution, considering the influence of the small scale of UMIST face datasets on the recognition rate, the data enhancement is used for training. The experimental results show that the method can effectively improve the recognition rate of the test image.