| Switch Filter is a conceptually simple but effective noise suppression filter. It has been demonstrated, by many researchers, that the switch filter can remove impulsive noise while preserving image details for greyscale image. In this work, this nonlinear filter is extended to process color image with both Mean Filter and Vector Median Filter incorporated in the framework for noise detection. The L*a*b* color space is selected as a perceptually uniform color space, which is a suitable model for Switch Filter to detect noise by comparing color difference. The conversion between the L*a*b* color space and the traditional RGB color space is nonlinear, which results in the dependence of the noise variance on the mean level. To preserve the property of noise, a variance stabilization transformation is derived. Extensive simulation studies demonstrate that such filter structure has a better performance than Sigma Filter when dealing with impulsive noise and a comparable output for the case of Gaussian noise. |
