An accelerated momentum based gradient projection method for image deblurring

Linear inverse problems (LIP) have been widely employed in wavelet-based image deblurring problems. The standard way is to minimize L2-L1 norm by gradient descent search in the solution space, which is typically suffered from poor convergence and unwanted local optima. In this paper, an accelerated image deblurring method is proposed by using momentum based gradient search scheme combined with the projected step technique. By adding a momentum term to the typical negative gradient direction, iterative steps will be effectively enlarged in favorable directions and some specific local optimal points can be avoided. Experiment results demonstrate that the proposed method outperforms its competitors, both in time efficiency and image restored quality.