Implementation and analysis of low power fast motion estimation using optimal binary encoding techniques based on hamming distances

Abstract

Motion estimation (ME) technique is an important feature in video compression. Block Matching (BM) is the popular technique for ME and it is utilized to reduce computational complexity. The block matching methods partition the frames into rectangular blocks and find the best motion vector (MV) with previous frames in search window. The unsuitable fast search algorithms occupy more memory space and more computational complexity for real time video coding applications. Every frame is partitioned into number of blocks with MV by using the Sum of Absolute Differences (SAD) or the Sum of Squared Differences (SSD) which computes the adjacent pixel blocks from current and reference frames. These techniques can be computed by the selection of search windows. Different types of methodologies have been introduced for finding best matching and best MVs. Three Step Search (TSS), Predictive Three Step Search (PTSS), Improved Three-Step Search (ITSS), Four Step Search (FSS), FSS genetic search, Diamond Search (DS), New Orthogonal Diamond Search Pattern(NODSP) , New Cross-Hexagon Search (NCHS), Cross Octagonal Search Algorithm (COSA), Adaptive Rood Search Pattern (ARSP) algorithms, Discrete Cosine Transform (DCT) and Redundant Discrete Wavelet Transform (RDWT) techniques have been utilized for finding best in matching and MVs. However some search pattern techniques failed to reduce the computational complexity, whereas some transform methods reduces only the memory bandwidth. Few search pattern techniques succeeded in reducing the computational complexity by overcoming its earlier search pattern techniques, but the overall accuracy performance gets degraded. To overcome all of the above issues, Optimal Binary Encoding Technique is proposed. An optimal binary encoding technique is based on binary code optimization of pixel values, which improves the overall accuracy performance in the motion estimation. The binary code optimization reduces the parallel memory resources in sub-pixel values and the optimized mapping function quan