Lossless Video Data Hiding for Privacy Protected Video Surveillance System (I)

Recently, video surveillance systems have become a part of our daily lives. For example: closed-circuit cameras in shopping malls have been widely used for crime detection, and employed to assess the charges for toll stations. However, the ubiquitous cameras, in combination with wireless transmission, and a strong recognition algorithm, are easier to monitor criminal acts, but on the other hand it also seriously damage the individual's privacy. Hence, how to use the techniques of data hiding in video is a trend for modern video surveillance systems to preserve individual's privacy. This kind of video surveillance system is named Privacy Protected Video Surveillance System (PPVSS). To develop a good PPVSS, one must consider whether it can be restored as cover video, and if the hidden video data (embedded video) has good visual quality, which cannot easily be detected. And the embedded video shall have high data hiding capacity and other characteristics. Basically, a bunch of video sequences is composed of many images (frames), and these images are highly correlated in spatial and temporal domain. So, in this project, we plan to use the property of the original video sequence to develop various data hiding algorithms for PPVSS. In summary, in the first year, this project will derive the difference image between the consequent frames, and apply a Side-Match VQ (SMVQ) method to produce relative codebook after the pixels for each encoded (encoding pixel). And, the compression, coding and data hiding methods are then developed. It is expected that each pixel in a frame can hide about 3 ~ 4 bits and the hidden data after removing the original video can be completely restored. In the second year, for the differences image obtained from the consequent frames, we plan to use the difference expansion and the histogram shift techniques to encode and hide information, so the embedded video is very similar to the original video in the visual quality. It is expected that each pixel of this method can hide about 1 ~ 2 bits, and the encoding time is significant reduced. This approach can be used for covert communication. At the last years, we will use the relationship of I, P, and B frames existed in MPEG-2 video, to establish the difference images for data hiding. In addition to using difference expansion and histogram shift techniques, the codebook-based coding is added to increase the amount of hidden data. Performance comparison with the above two schemes developed in previous will be investigated. Upon completion of this project, It is expected that the proposed algorithms will promote the development of PPVSS.

Project ID:PB10007-7260
External Project ID:NSC100-2221-E182-047-MY3