Scalable motion vector coding for MC-EZBC

For a scalable video coder to remain efficient over a wide range of bitrates, the motion information should be represented in a scalable manner. Indeed, especially at low bitrates tradeoffs should be made between motion information representation and texture information accuracy. Consequently, motion vector scalability has emerged as an important research topic, in conjunction with fully scalable video compression schemes. In this paper, a new scheme providing motion vector scalability is proposed. Starting from the MC-EZBC scalable wavelet video coder [1], multiple motion layers are constructed according to the importance of motion information. The overhead associated with this layered representation is proven to be negligible. Hence, the rate-distortion performance at high bit-rates is about the same as that of the scheme without motion information scalability, while at low bitrates the scheme benefits from this scalable coding strategy.