Joint color and texture descriptor using ring decomposition for robust video copy detection in large databases

Abstract

The growth of audiovisual content, and in particular video requires the creation of robust tools for detecting illegal copies. This paper presents an effective approach to search and detect illegal copies in large video databases. This automatic detection operates two local descriptors and their paths through the video. This method allows to reduce the temporal redundancy intrinsically linked to the video and to add a context of behavior in these descriptors. Thus, starting from a low-level signal description, our approach allows to achieve a higher level of representation for two reasons; i) it uses a new way to decompose video frames based on a ring decomposition and, ii) it combines a local texture descriptor namely binarized statistical image features (BSIF) extracted using a ring decomposition and local color descriptor. The obtained description is more compact, non-redundant and can be highly robust to rotation and flipping. The evaluation shows a clear improvement in performance against other novel techniques listed in the state of the art while exhibiting better flexibility. It is more real time on a large video base (TRECVID 2009).