Image sequence inpainting: Towards numerical software for detection and removal of local missing data via motion estimation

摘要

Film restoration aims to remove or reduce various types of film and video defects in order to provide visual enhancements of image sequences. The automatic treatment of these defects is a challenge. Restoration is still performed by hand even if by using numerical techniques for retouching. This is a very intensive activity and great improvements, both in quality and in speed, can be obtained by using automatic or semiautomatic software.This paper surveys the overall computational steps needed for the development of effective software tools to be actually used in a concrete application. In particular, here we focus on recovery and reconstruction of a particular local random defect of old black-and-white films, commonly referred to as “blotch”. We start from the characterization of the degradation model both for detecting and for restoring the defect and deal with such inverse and ill-posed problem through edge preserving regularization. We employ a spatio-temporal interpolation for blotch removal where the initial approximation is given by interpolating along the motion trajectory data belonging to adjacent frames. Finally, we describe the numerical algorithm and some experimental results.