Image Denoising Using Information Measure and Support Vector Machines

Abstract

Image denoising is one of important steps in a number of image processing applications. However, available methods mainly present by conducting filter of restoration on whole observation image, resulting in many image detail information have been lost. So, how to obtain the balance of remove noises from the smooth regions and preserved more image detail at high frequency regions have still worth to pay more attention. It is presents a novel approaches that can improve image quality by reducing corrupted pixels, but leave good pixels unchanged. First, information measure method is introduced to extract noise features from observation image. And then, a support vector machines (SVM) based classifier which is employed to divided noise corrupted image into noise candidates pixels and good pixels, so that a noise map is generated that can be used to guide the Mixed Mean and Media Filter (MMMF), which is designed to conduct restoration filter just for corrupted pixels. Three typical numerical experimental are reported and results show that the proposed algorithm can achieve better performance both on vision effect and a higher mark on objective criterion(Peak Signal and Noise Ratio, PSNR).