A review of video registration methods for watermark detection in digital cinema applications

To deter constant theft in digital cinema applications, forensic watermarking has been proposed to track a digital content and to determine the time and the place where the theft occurs. However, to be effective in digital cinema applications, a forensic watermark must be invisible and secure. The detection of the forensic watermark must also be robust to attacks and processing, such as camcorder capture, low bit rate compression, D/A/D conversion, cropping, resizing and frame rate conversion. To achieve the required robustness detection, many proposed forensic watermarking algorithms adopt reference-based (informed or nonblind) watermark detection where the original video is used to assist the detection. To use the original video, one must first register it to the candidate video, where the forensic watermark needs to be extracted. We will review a number of video registration algorithms proposed for forensic watermark detection in digital cinema applications. Among all the proposed registration algorithms, we will discuss and compare the following three algorithms: (1) a semiautomatic temporal, spatial and volumetric video registration algorithm proposed by IRISA, (2) a key frame based temporal alignment algorithm proposed by Universite Catholique de Louvain and (3) a spatial, temporal and volumetric video registration algorithm proposed by Sarnoff.