A multistage video shot boundary detection approach: robust to motion and illumination effects

Abstract

In this today’s modern era because of increasing digitization and excess growth of multimedia data, video indexing and retrieval tool have become very essential in the field of video retrieval based on content, where video shot boundary detection is regarded as a stage of develop-ment for sudden changes in motion and brightness. A robust multi stage Shot Boundary Detection (SBD) technique is proposed in this present work using Uniform LBP, Canny Edge Detector and Lab Color Differ-ence. The proposed system can identify abrupt shot transition in a video, which are invariant to motion and intensity effects. For extracting fea-tures from each of the frames of a video uniform Local Binary Patterns (LBP) is used. Adaptive threshold δ1 is used to identify possible tran-sition frames. Adaptive threshold δ2 and LAB color difference is used to extract and identify actual transition frames. Motion effects and illu-mination effects are handled in two different phases. At the first stage, effects of brightness and motion effects are reduced by Canny edge de-tection and in the second stage, Lab color difference is used to handle illumination effects. TRECVid 2001 and TRECVid 2007 datasets are used for observation purposes. The proposed method provides good F 1 score compared to other recent shot boundary detection approaches.