On the Suppression of Mixed Gaussian Impulsive Noise in Color Images

In the paper the efficiency of a robust denoising technique whose aim is the suppression of mixed Gaussian and impulsive noise in color images is analyzed. The proposed filtering framework is based on a novel concept of closeness between a sample from a processing block and the local neighborhood (patch) of the pixel, which is being restored. The dissimilarity measure is defined as a sum of squared distances to the most similar pixels of a patch centered at the processed pixel. The filter output is a weighted average of samples from the filtering block and the similarities to the central patch are used as weighting coefficients. To increase the filter robustness, the impulsiveness of pixels in the processing block is also taken into account, which significantly improves the filtering performance.The novel technique has been compared with the widely used filtering methods and the performed experiments proved its satisfying smoothing properties. The new filter is insensitive to outliers introduced by the impulsive noise process, enhances image edges and preserves details, while efficiently suppressing the Gaussian noise component.