Intelligent Image Filtering Using Rough Sets

Abstract

The paper presents a class of novel, high-quality image filters which are based on the rough sets and which both effectively remove noise and enhance edges. Current filtering techniques do not do both effectively: some enhance edges but do not remove noise sufficiently, and most filters blur edges and small image details when removing noise. Many filtering techniques lose the shape details because of statistical averaging or sorting gray levels within a window. The novel methodology presented in this paper uses the upper approximation to check good continuation of the window center with templates distributed uniformly around the center. A non-statistical variance between the center of a window and pixels of window templates is used as the measure of good continuation. The minimum value of this semi-variance is found to get the most homogeneous template. The average or median of this most homogeneous template is then used as the gray level of the window center and is assigned to the pixel gray level in the filtered image. In addition, adaptation of template shape to the region shape in the image is accomplished. Several variations of the technique have been constructed on the top of other filters.