Research On The Motion Estimation Algorithms Based On The Redundant Discrete Wavelet Transform

Block-based motion estimation in the spatial domain is widely employed in modern video compression systems. However, given the promising performance of wavelet-based still-image compression algorithms, now one hot spot in the research of video codec is the motion estimation algorithm based on the Discrete Wavelet Transform (DWT). However, the fact that the usual critically sampled DWT is shift variant greatly hinders the motion estimation process when deployed in the wavelet domain. Because the Redundant Discrete Wavelet Transform (RDWT) is shift invariant, the motion estimation algorithms in the redundant wavelet domain have good performance, but very high computational complexity.This paper mainly researches on the RDWT-based motion estimation algorithms. Because of the high computational complexity of such algorithms, so based on the analysis of the motion characters of video sequences and the existing motion estimation algorithms, utilizing the characteristic of RDWT, this paper proposes a fast multi-resolution motion estimation algorithm based on the RDWT. Firstly, this method by identifying the potential motion blocks and dividing the blocks into different degrees of motion in the RDWT domain takes different search schemes to blocks of different types, so the search area is decreased efficiently. In addition, this paper introduces the multi-resolution motion estimation into the redundant wavelet domain. Motion estimation first happens in the low-band, and then the motion vectors are refined in the high-band with a smaller search area. This motion estimation algorithm gets a tradeoff between the precision and the computational complexity, and overcomes the high computational complexity of the existing RDWT-based motion estimation algorithms.The experimental results prove that this approach has a superior performance compared with the traditional method in terms of the precision and the computational complexity.