Impulse Noise Removal For Color Image

Recently color image filters have been used widely in image processing, computer vision, telecommunications and so on. Because of sensor malfunction, imperfect optics, electronic interference, or flaws in data transmission procedure, impulse noise will be introduced into color images, which affects the perceptual quality of the image decreasing the performance of the task for which the image was intended. Therefore, filtering is an essential part of any image processing system.Color image filters can be generally divided into two classes: linear and nonlinear. Compared with linear filters, nonlinear filters could preserve edges and details better while denoising, which avoids blurring and details loss. Vector median filter (VMF) is a common and effective nonlinear filter, but it treats all pixels equally, many high-frequency signals are filtered out which results in image blurring.This paper presents three novel nonlinear filters with a two-step framework of impulse detection and noise removal. The first is kernel density estimation (KDE) detection based random walk ordering filter. This method applies kernel density function and uses it to detect impulses. It introduces random walk concept into vector ordering and finds the output which maximizes transition probability from central pixel to its neighbors. The second is progressive switching median (PSM) detection based VMF. It computes the difference of pixels to find the impulse. VMF computes median value of uncorrupted neighbors as its output. The third is rank-ordered absolute differences (ROAD) detection based exponential similarity ordering filter. ROAD is used to determine whether a pixel is impulsive to be removed. It does noise suppression based on maximizing the proposed pixel similarity measure that considers both magnitude and angular information of pixel vectors'difference.Many standard images corrupted by various percentage of impulses are filtered by new approaches, MAE, MSE, NCD are used to evaluate the performance of filters. Experimental results show that our new approaches achieve good results in impulse removal and detail preserving.