Evaluation of a mosaic based approach to video compression

There is great interest in the storage and browsing of large collections of video clips, due to the current explosion in demand for digital video. Many new algorithms and standards (MPEG, H.261, H.263, etc.) have been developed in recent years for the coding of video. However, most of these techniques lose performance when there exists significant camera motion, a common element of some classes of video such as surveillance and sporting events. A mosaic-based approach to video compression, initially presented by Irani, Hsu, and Anandan (1995), creates a world-oriented panoramic representation of the scene for compression. This paper proposes and evaluates a specific mosaic-based algorithm for video compression that uses a static mosaic with a wavelet-based coding scheme for the mosaic and the temporal residual information. The mosaic-based approach is compared to a global motion compensation approach and the algorithms detailed in this paper are compared to the H.263 TMN5 coder