A Switching vector median filter for impulse noise removal from color images

A new filter for removing impulse noise from color images is presented which is based on the modification of median of absolute deviations from the median (MAD). Quaternion theory is used for expressing the color difference between two pixels. It considers both the intensity and chromaticity differences. The proposed algorithm is a two-stage scheme. In the first stage, an impulse noise detector is developed to discriminate the color pixels which are likely corrupted by noise. The image is enhanced in the second stage by replacing the noisy pixels with the output of the commonly used vector median filter computed in quaternion form. The obtained results show that the proposed filter achieves good filtering performance over other existing filters. It integrates the advantages of quaternion expression of color images which preserve vector nature by treating it as single entity and robust impulse detection by MAD.