Automic redundant event detection for video

In this dissertation, a new methodology has been proposed to determine retake in rushes video. In this methodology, the video is divided into shots by the proposed automatic Shot Boundary Detection (SBD), which uses local Singular Value Decomposition (SVD) and k-means clustering. Shots that contain a single color, color bars or clapper boards will be eliminated by our proposed algorithm and Near-Duplicated Keyframe (NDK). In the remaining shots, the local features of each frame are extracted using Scale-Invariant Feature Transform (SIFT) algorithm. The similarity between consecutive frames is calculated using a SIFT matching and then converted into a string. The given string is then concatenated into a string sequence to use as a shot representative. The similarity between two sequences is evaluated by the Longest Common Subsequence algorithm (LCS). In the experiment, first, our automatic shot boundary detection is compared with conventional technique. Second, results of retake shots are compared with results from conventional technique. Results show that our proposed methodology provides a reasonably high degree of accuracy to detect a retake in rushes video.