Detection of line scratch using energy regularity of spatio-temporal video cube

Scratch is the most dominant defect in an old age video film and its most common form is the vertical line scratch. Restoration of video film corrupted by these defects is totally depends on their absolute detection. Generally, line scratch is persistent in a sequence of frames and their pattern is almost regular as compared to the other defects. Detection of regular pattern is easy as compared to the irregular pattern and this is the key point of the proposed detection method. In the proposed method, non-traditional TY and TX frames has been used rather than a traditional XY frame of video. It is a non traditional approach of line scratch detection method. In the proposed method, modelling of video is formulated by using spatio-temporal regularity flow. Spatio-Temporal regularity flow (SPREF) framework is used here for video modelling, which assumes video as cube and process whole video cube at a time. The pixel intensity varies the least for regular pattern and irregularity produces large variation. Such irregularity is detected by the energy of the frame. The cause of irregularity is not only due to the large intensity variation, but also due to the deviation of flow vectors. Energy of frames incorporates both the intensity and flow vectors of the frame. Experiments have been formulated on the standard test videos and achieved higher detection accuracy that proves its robustness.