Content fingerprinting from an industry perspective

Content recognition technology, particularly digital fingerprinting of audio and video content, has important applications in the distribution of movies and television programming on the Internet. However, the identification process is complex as distributed content is often substantially altered prior to distribution through processes such as cropping, shape transformations, noise, signal distortion and transcoding that make recognition difficult. Fingerprinting has emerged as a robust and general approach to content identification. MovieLabs and the Motion Picture Association of America have been conducting tests of various fingerprint algorithms against real-world content. Our findings indicate that the best algorithms can achieve match accuracy greater than 98% with a 0% misrecognition (false positive) rate against severely transformed content. This paper describes applications of content recognition, fingerprinting, criteria and methods for evaluating fingerprinting, and describes our fingerprint testing.