MULTISTAGE ORDER STATISTIC FILTERS FOR IMAGE SEQUENCE PROCESSING

There is a vast number of applications in medical imaging, computer vision, and video communications, where the processing of image sequences is critical. Although many techniques have been successfully developed for tasks such as 3-D motion estimation and image sequence coding, very few techniques have been proposed for the restoration of noisy image sequences. This is due, mainly, to the artifacts introduced when temporal filtering is applied to video containing moving objects. In this paper, we study the application of multistage order statistic filters (MOS) for the task of noise suppression in time-varying imagery. It is shown that MOS filters efficiently preserve image structures under motion without motion compensation preprocessing. In particular, we consider the families of multistage median and weighted median filters. Motion preservation and statistical smoothing measures are derived. Finally, it is shown that spatiotemporal filtering allows for a significant improvement over both spatial and temporal filtering in terms of output image resolution and noise suppression.