Adaptive order statistic filters: the complexity/quality tradeoff

Abstract

The authors compare, in some detail, six adaptive order-statistic-based filters with the median filter for image processing purposes. The most commonly used order-statistic filter is the median filter since it is easy to implement and removes impulse noise while preserving edges. One problem of the median filter is that its fixed window size constrains its performance. A large window size will give good impulse noise suppression but may blur the image while a small window size may not adequately remove the noise. Another problem is that the median filter is not the optimum filter for removing Gaussian noise. Each of the six adaptive order-statistic filters examined attempts to solve one or both of these problems, with the tradeoff being increased computational complexity for better image quality. When choosing a filter one must look at the computational complexity, the type of noise to be removed, the image quality required, and what kind of prior knowledge is required by the filters. The seven filters are examined for a variety of images and noise types. Some image quality results are presented. >