Comparisons and Analysis of Motion Estimation Search Algorithms Ihab Amer, Karim Abul-Ainine German University in Cairo (GUC) Main Entrance, Fifth Settlement, Cairo, Egypt ihab.amer@guc.edu.eg karim.abdelhamid@student.guc.edu.eg Wael Badawy, Graham Jullien ATIPS Laboratories ECE Dept. University of Calgary 2500 University Dr. NW, Calgary, AB {badawy, jullien}@atips.cs Abstract Motion Estimation/Compensation is one of the most critical modules in a typical digital video encoder. Many implementation tradeoffs should be considered while designing such a module. This paper focuses on the implementation of several motion estimation algorithms and investigates the differences between them according to different evaluation metrics. The algorithms are evaluated with respect to different parameters and are applied on different video sequences with different resolutions and types of motion. 1. Introduction In a video sequence, successive frames typically exhibit a high degree of correlation. Such phenomenon results in the temporal type of redundancy in a video signal. Interprediction creates a model for one or more previously encoded frames to get rid of such redundancy. One way to do that is by a process called motion estimation, where each block in a frame is represented by a motion vector that represents the displacement of the block with respect to another “highly correlated” block in a previously encoded frame [1]-[3]. Many motion estimation algorithms have been proposed in the literature, ranging from the typically impractical brutal-force full-search motion estimation algorithms up till the wide variety of more practical fast-search algorithms. Applying fast searches typically return relatively lower quality than the full search algorithm. However the drastic savings in terms of computational requirements strongly favors fast searches for many applications [2]. The remainder of this paper is organized as follows: Section 2 provides a brief description about the algorithms under consideration. Section 3 contains the results and analysis. Finally, section 4 concludes the paper. 2. Algorithms Description This section provides a brief description of the implemented motion estimation search algorithms. The section starts with the full search, and then it moves to various fast search algorithms. 2.1. Full Search The full search technique is very systematic, but on the other hand, it is impractical in terms of computational complexity/runtime. For each block in a current frame, every candidate block within the search window in the referenced frame is searched. The motion vector is then calculated pointing to the block with the lowest distortion in the reference frame. Figure 1 shows two different approaches used in the full search procedure. (a) (b) Figure 1 - Full search algorithms (a) Spiral (b) Raster 2009 World Congress on Computer Science and Information Engineering 978-0-7695-3507-4/08 $25.00 © 2008 IEEE DOI 10.1109/CSIE.2009.229 601 2009 World Congress on Computer Science and Information Engineering 978-0-7695-3507-4/08 $25.00 © 2008 IEEE DOI 10.1109/CSIE.2009.229 601