Correlation Based Rood Pattern Search (CBRPS) for Motion Estimation in Video Processing

Abstract

To exploit the temporal and spatial redundancy and to expedite the process of video encoding it is required to find the motion Vector (MV) by a process called Motion Estimation (ME). In this paper, optimal motion estimation algorithms based on correlation that exists between consecutive macro blocks in a frame of a video with stopping criteria are proposed. The different neighborhood macro blocks (MB) form the Locality of Reference (LOR) of the candidate macro blocks in the reference frame and these LOR defines the types of algorithms which are named as Correlation Based Rood Pattern Search (CBRPS) in four variations which effectively reduce the search time, computational complexity without compromising the PSNR values i.e. quality of the compensated image. The results obtained are compared with the results with relevant existing techniques to highlight the superior performance of the proposed algorithms. All the proposed algorithms outperform the existing techniques both in terms of computational complexity and PSNR values. Amongst the four proposed algorithms the fourth type of CBRPS i.e. CBRPS_T4 is found to be the best for linear and transverse motion of objects in the videos in terms of computational complexity.