A fuzzy filter for the removal of random impulse noise in colour video

In this paper a new filtering framework for colour image sequences corrupted by random impulse noise is introduced. The proposed method consists of three successive filtering steps in order to find a good trade-off between detail preservation and noise removal. One hard filtering step, that should remove all the noise at once, would namely also remove a considerable amount of details. In the different noise detection steps, we make use of human knowledge represented in the form of fuzzy if-then rules. The detection of noisy pixel components is based on spatial and temporal information as well as on information from the other colour bands. Additionally, only detected pixel components are filtered by blockmatching based on an adaptation of the mean absolute difference (MAD) to noise. The other components remain unchanged. From the experimental results, it can be concluded that the proposed filter outperforms other state-of-the-art impulse noise filters.