Non-Local Directional-Guided Filter for Impulse-Gaussian Mixed Noise Image Denoising

We introduce an effective technique to restore the images corrupted by additive Gaussian noise and impulse Salt and Pepper noise. In this Work, a three-step non-local directional-guided filter is seted up. We begin by identifying Salt and Pepper noise, estimate intensity of mixed noise and preliminarily remove and repair it by Maximum Likelihood Estimator. Afterwards, use a set of discrete total variation (TV) models to mine potential directional information and generate a set of directional-guided templates. At last, We build a non-local directional-guided filter to restore lost details. Experimental results verify that the proposed algorithm can obtain the best denoising performance compared With some typical methods. In the case of high intensity noise pollution, our algorithm has more advantages.