Two pixel-preselection methods for median-type filtering

Most of the median-type filters designed to remove impulsive noise process all the pixels in the noisy image. A certain number of non-noisy pixels are thus unexpectedly modified, thereby limiting the quality of the result. Since these pixels need not, and should not, be filtered, the authors propose two approaches for selecting the pixels to be processed. The first of these operates online, by relaxing the order statistic for pixel substitution, this is known as the relaxed median filter. The second approach requires the construction of a mask to determine a priori which pixels are to be processed, this is known as the mask-median filter. The mask is built using some prior information about the class of images at hand