Simultaneous Blurred Face Restoration And Recognition

Blur is one of the most common forms of image degradations.For blurred faces,usually two tasks are performed,that is,blurred face recognition and blurred face restoration.So far various algorithms have been proposed for either one of the tasks,yet only a few consider them both.This paper analysizes the connection between blurred face restoration and recognition.For blurred face restoration,we propose two new blurred face recognition-based models,which could handle the two problems existing in the state-of-the-art exemplar-based face deblurring method.In model 1,we apply the linear representation of the training dictionary to solve the intraclass varia-tion problem.In model 2,we use L0.8 prior instead of LO prior on the gradient image for a more proper sharp pure face region.Then the two models are com-bined together to form a Two-Step solution for our blurred face recognition-based restoration problem.For blurred face recognition,we compare the blur-ring method and deblurring method and finally choose the better-performing blurring method combining with LPQ feature.At last,we build our Simulta-neous Blurred Face Restoration and Recognition(SRR)algorithm by iteratively solving the two tasks.The SRR method is a good choice to deblur face images with complex blur kernels as well as undersampled blurred face recognition problem.Experiments on FERET database demonstrate the efficiency of SRR in terms of restoration qualities and recognition accuracy.