Motion estimation using multi-resolution based on Haar wavelet transformation

Abstract

Out of several block matching motion estimation techniques, full search block matching (FSBM) motion estimation is the best so far when the quality of the video is concerned. However, the FSBM algorithm is computationally expensive and hence not suitable for real life video. On the other hand, the three step search (TSS) algorithm requires lesser number of computations and suitable for low bit rate video applications. The TSS algorithm requires complex hardware compare to its FSBM counterparts. In this paper a new motion estimation techniques using the concept of multi-resolution is proposed. Our proposed motion estimation technique follows the three steps in the TSS algorithm, but each step has a different image resolution. To achieve the different resolutions of an image, we use Haar wavelet transformation. It has been experimented that the proposed technique requires even lesser number of computations compared to the TSS algorithm. So when it comes to video quality, the proposed algorithm outperforms the TSS algorithm. Further, the algorithm is synthesizable to a simple VLSI architecture with lesser number of gate counts and lower power dissipation.