Line scratch detection using Spatio-Temporal regularity flow vectors

Vertical and horizontal line scratches are primarily defects in old video data recorded on film. The flawless restoration of scratches is totally dependent on their perfect detection. False and missed detection weakens the restoration result. In this paper we address the problem of detection of primary and secondary line scratches and also propose a solution for false and missed detection. The problem of simultaneous detection of both dark and bright intensity scratch is also solved by proposed method. The proposed detection method is based on selecting non-traditional TY and TX frames rather than the traditional XY frames of a video. Unlike prior detection techniques that are based on frame by frame processing of XY frames, the proposed method of detection uses a video cube consisting of a group of TY or TX frames and exploits the Spatio-Temporal feature called the regularity flow in a cube. The regular flow of a pixel in a cube indicates the absence of a scratch while irregularity in the flow indicates the presence of a scratch. Experimental results show that the proposed method has no false detection for line scratches of width 1 to 10 pixel.