F-SURF Feature Descriptor for Video Copy Detection

摘要

Video Copy Detection focuses on preventing illicit use of digital videos. Video copies are generated by applying different sorts of transformations on the original video content. To detect such transformed copies, the extraction of a transformation invariant feature descriptor is a requisite. Among the various existing transformations, flipping is the recently employed copy attacks. Hence, we propose a flip invariant feature descriptor named F-SURF (Flip invariant SURF) for extracting transformation invariant feature content. F-SURF achieves flip invariance in SURF feature descriptor. In this paper, a novel Stochastic Dimensionality Reduction approach is also proposed which reduces the computational complexity associated with F-SURF descriptor by performing a dimensionality reduction technique. Video copy sub sequences are identified using a hierarchical approach, which effectively determines the matching copied segments. Experimental analysis reveals that the system has a retrieval accuracy of 85%.