Due to the capture process, video signals are generally contaminated by noise. Furthermore, overall video quality can be improved by reducing sharpness in fast moving areas, except in cases where the eye is able to track the motion. In this paper we propose a relatively simple pre-processing method that is able to address both these situations. The proposed pre-processing algorithm is based on selectively removing high frequencies that are not well predicted by the motion compensation. While the decision is made based on temporal tracking, the filtering is done exclusively in the spatial domain, thus avoiding the artifacts produced by other pre-processing methods.