Enhanced Efficient Diamond Search Algorithm for Fast Block Motion Estimation

Abstract

Although the conventional FS algorithm achieves the best quality amongst various Motion Estimation (ME) algorithms and it is straightforward and has been successfully implemented on VLSI chips [1], its computational complexity is very high.In contrast, real time and portable multimedia devices require ultra computationally efficient video codec designs that will allow for a robust and reliable video quality.Many sub-optimal but faster ME t e c h n i q u e s h a v e b e e n p r o p o s e d t o t a c k le the previous computational complexity problem.One technique is based on simplifying or easing the matching criteria (SAD) based on spatial and/or temporal Macro Blocks' (MB) features.Partial Distortion Elimination algorithm (PDE) and Successive Elimination Algorithm (SEA) [2] are examples of such techniques.Another technique is based on reducing the number of search points in the search area [3].Although this technique reduces the computational complexity, there will be degradation in PSNR.New Three Step Search (N3SS) [4], Four Step Search (FSS) [5], Predictive Motion Vector Field Adaptive Search Technique (PMVFAST) [6], Hexagon Based Search (HEXBS) [7], Diamond Search (DS) [8], and Cross Diamond Search (CDS) [9] are examples for such techniques.In this chapter, a new simple and efficient fast MDS is developed for higher complexity reduction than DS without further PSNR and bit-rate degradation compared to FS. DS is the most accurate suboptimal ME algorithm among others.This is why it was chosen to be implemented in the reference software of the standard H.264.The proposed MDS algorithm uses a mixed flavor approach of the previous two techniques.A computationally efficient set of stop search algorithms that skip the unnecessary operations both internally within the MB and externally between MBs utilizing accurate adaptive threshold models for reducing computations with no significant degradation in PSNR compared to the FS algorithm.Internal SAD operations are minimized using DISS.DESS algorithm is used to eliminate the irrelevant candidate pixels in the search area.Moreover, we are proposing to enhance the mechanism of the DS algorithm using zero and center bias properties.A set of adaptive and accurate thresholds that early terminate the search or select between Small Diamond Search Pattern (SDSP) or Large Diamond Search Pattern (LDSP) results in an additional computational saving.It is worth mentioning that some related work of this chapter was published in [11].Coming up in section 2, details on the proposed modified search algorithm are explained.Following that, in section 3 comparisons are made between the proposed MDS and other fast motion estimation algorithms.Finally, we draw conclusion in section 4.www.intechopen.com