Comparison of distance measures for video copy detection

Abstract

Content-based copy detection (CBCD) is a complementary approach to watermarking for detecting copies of media. Watermarking relies on the ability to detect from a copy a distinct pattern that was introduced into the original media. CBCD techniques detect copies by measuring distances between content-based signatures extracted from the original and the copy. The critical challenge in content-based copy detection is the design of signatures, which are invariant to the differences in quality across copies of the same media (resolution, compression and digitization effects). Most of the distance measures used in image retrieval have been developed without much consideration to these types of variations between copies. This paper examines the use of several image distance measures in the context of video copy detection and compares their performances.