Detection of dirt impairments from archived film sequences: survey and evaluations

Abstract

----Film dirt is the most commonly encountered artefacts in archive restoration applications. Since dirt usually appears as temporal impulsive event, motion-compensated interframe processing has been widely applied for its detection. However, motion-compensated prediction requires a high degree of complexity and can be unreliable when motion estimation fails. Consequently, many techniques using spatial or spatio-temporal filtering without motion have also been proposed as alternatives. In this paper, a comprehensive survey and evaluation of existing methods is presented, in which both qualitative and quantitative performances are compared in terms of accuracy, robustness and complexity. After analysing these algorithms and identifying their limitations, we conclude with guidance in choosing from these algorithms and also promising directions for future research.