Edge preserving smoothing

Edge preserving smoothing is an image processing technique where the edge information is preserved during the smoothing process. It uses non-linear operator which is able to remove texture and noise, while keeping edges and corners. The technique is known also as edge and corner preserving smoother (ECPS).

General

Smoothing is an important task in image processing. The best known smoothing technique is low-pass linear filtering. The most widely used filter deploys a Gaussian function as a smoothing kernel. However, since linear low-pass filtering strongly attenuates high frequency components, not only noise, but also edges and corners, are smoothed. Therefore, there has been a remarkable effort to develop nonlinear operators able to remove texture and noise while preserving edges and corners. In the following we refer to such an operator as an Edge and Corner Preserving Smoother (ECPS).Several ECPSs have been proposed in the literature. The best known ones are based on median filtering , morphological analysis, bilateral filtering, mean shift, total variation, and anisotropic diffusion. The latter is probably the most popular ECPS, for which much research has been carried out in the last fifteen years. However, it is not computationally efficient since it requires many iterations to achieve the desired output.

Edge preserving smoothing for artistic imaging

An important aspect of ECPSs is their ability to produce images that are visually similar to paintings. Not all existing ECPSs are suitable for producing such an artistic effect. For this purpose, an interesting class of ECPSs stems from the early work of Kuwahara [21] , where a fast and conceptually simple ECPS is introduced. A symmetric square neighborhood around each pixel is divided in four square windows. The value of the central pixel is replaced by the gray level average over the most homogeneous window, i.e. the window with the lowest standard deviation. Although this operator was not specifically designed for producing artistic images the obtained effects are quite interesting. Suffessive development of this filter concerns the value and criterion filter structure, which include the Gauss Kuwahara operator as an important sub-case.

Comparison of different techniques for EPCS, with application to artistic imaging, can be found on http://www.cs.rug.nl/~imaging/artisticsmoothing

ee also

* Non-photorealistic rendering
* Art software