Robust video analysis: Which statistics to use?

The aim of this paper is to discuss which statistics are efficient when video sequences are highly corrupted by impulsive noise. Usually the Gaussian model is implicitly used for modelling the noise. Many papers in the literature use the Gaussian mixture as an alternative of this later for image and video processing (J.C. Brailean et al., 1995). No a priori assumption for the noise pdf will be considered. We will use higher-order statistics to improve the processing results instead of second order ones used in the Gaussian case. Experimentations show the effectiveness of the proposed approach.