One-dimensional full search motion estimation algorithm for video coding

Abstract

Abshct-A new hardware-oriented algorithm called the onedimensional full search (1DFS) is presented for block-matching motion estimation in video compression. The simulation for this algorithm fdlows A.261 and MPEG international standards. In MPEG simulation, stroetures with 1-, 2-and 3-frame interpolation are compared. The performance of lDFS is superior to that of other fast search algorithms. And it has more regular data flow, data r e w and less control overhead. It is an alternative for 2D full search block matching and achieves a good compromise between computational complexity and performance. With competent performance and " a b l e computation complexity, the proposed method is more suitable for real-time hardware realization of a W I motion estimator for video applications. I. INTRODUC~ONIn interframe predictive digital video coding system, the temporal redundancy due to high correlation between consecutive frames can be eliminated by motion estimatiodcompensation technique. There are many types of motion estimation algorithms such as pel-recursive, block-matching [ 11 and feature-based approaches. In general, the block-matching algorithm (BMA) is dominant and more suitable for a simple hardware realization. Most of current video communication systems and standards employ a block-matching scheme for motion estimation (such as CCI'IT proposed H.261 (p x 64) standard [2] and MPEG (Moving Picture Experts Group) [3] full-motion standard).A straightforward BMA is the two-dimensional full search algorithm (2DFS). 2DFS matches every possible displaced candidate block within the search range in the reference frame in order to find the block with minimum distortion. However, the implementation of 2DFS always needs massive computational efforts and huge hardware cost to achieve significant speed. Several fast BMAs have been developed in the past to alleviate the heavy computations of 2DFS, such as three-step hierarchical search (3SHS) [4], [lo], 2-D logarithmic search (LOGS) [5], one-at-a-time search (OTS) [6], [7], parallel hierarchical one-dimensional search (PHODS) [ 121, etc.OTS is one of the incremental fast algorithms that can provide quick computation and reasonable performance. However, considering the hardware implementation, the variable computation steps and searching directions will limit its applications. The experiments in [12] had also shown the high possibility that OTS can be trapped by local minimum. [I21 proposed a PHODS algorithm to solve the problem. Basically, PHODS combines the concept of hierarchical search and parallel computing to get good performance and easy hardware realization. But, the inherent drawbacks of hierarchical search finally eliminate the performance improvement of PHODS. Therefore, it is worth to find an algorithm to solve these problems.

Hierarchical adaptive rood pattern search for motion estimation at video sequence analysis

“…Из-за этого недостатка начали искать другие эффек-тивные алгоритмы. Многие варианты таких алгоритмов [2][3][4][5][6][7][8][9][10] были разработаны для оценки движения на основе блока. В настоящей работе рассмотрим такие алгоритмы быстрой оценки движения, как пол-ный перебор (FS), трехшаговый поиск (TSS), четырехшаговый поиск (FSS), ромбический поиск (DS), перекрестный ромбический поиск (CDS) и адаптивный метод шаблонного поиска (ARPS).…”

Survey on Block Matching Motion Estimation Algorithms and Architectures with New Results

“…Roughly speaking, a newer algorithm requires less computation, achieves faster convergence, and results in higher video quality. In [12,13], and [24], not only the number of searched candidates but also the feasibility of parallel processing, regularity of data flow, and efficiency of memory access were taken into consideration toward system optimization. The first diamond search [26], which is adopted by the reference software of MPEG-4, has significantly better performance in speed and quality than its prior algorithms.…”

“…Since 1981 many algorithms of this type, e.g. two dimensional logarithmic search [7], three step search [8], conjugate direction search [9], modified logarithmic search [10], cross search [11], parallel hierarchical one dimensional search [12], one dimensional full search [13], new three step search [14], four step search [15], block-based gradient descent search [16], centerbiased diamond search [17,18], advanced diamond zonal search [19,20], minimum bounded area search [21], one-dimensional gradient descent search [22], cross diamond search [23], predictive line search [24,25], and many others, have been proposed. Roughly speaking, a newer algorithm requires less computation, achieves faster convergence, and results in higher video quality.…”

A novel low-power full-search block-matching motion-estimation design for H.263+

“…For real-time application, either dedicated hardware or a fast algorithm should be applied. Nowadays, there are many fast algorithms for block-matching motion estimation, such as a three-step hierarchical search [2], 1-D full search [3], 2-D logarithmic search [4], clustering search [5], and sub-sampled search [6] … etc., but the degradation of the picture quality and the irregular data flow make it less attractive in system view. On the other hand, due to the amazing progress of IC fabrication technique, a full-search algorithm could be efficiently implemented in a video system.…”