Abstract

Face Recognition performance due to effect of blur in uncontrolled environments is still a challenging task. The image quality degraded by blur corrupts the face information, which further has a great impact on recognition rate. In many cases, blurred query face is deblurred and the original database is blurred and deblurred for efficient face recognition. But this increases the computational complexity and memory requirements. The method proposed here uses different local descriptors of the original database and local descriptors of the deblurred query face. Basically, by considering natural images are sparse in some domain, an Adaptive Sparse Domain Regularization method is used to deblur the Gaussian blurred face images. As this image restoration method consists of two adaptive regularization terms, the face image details can be constructed in terms of visual perception and PSNR. Further extracting distinctive features of the face image, reduces the complexity in the image scale. AR database is used in this experiment. Different feature descriptors such as Local Binary Pattern (LBP), Local Phase Quantization (LPQ), Histogram of Oriented Gradient (HOG), Local Gabor Binary Pattern (LGRP) are taken for the Gaussian deblurred face images and original dataset. Finally, Linear Collaborative Discriminant Regression Classification (LCDRC) is used for recognition by considering these feature descriptors of original dataset for training and feature descriptors of Gaussian deblurred dataset for testing. AR database is used in this experiment. The best recognition rate of 96.6% is obtained for Gaussian blurred faces using the proposed method.

Paper Details