MATLAB Implementation of EPZS Motion Estimation in H.264 AVC

Abstract

Abbreviations: EPZS, enhanced predictive zonal search; ME, motion estimation; FS, full search; MC, motion compensation; BMA, block matching algorithm; SA, search area; SAD, sum of absolute difference; AVC, advanced video coder; DS, diamond search; ES, exhaustive search; MVFAST, motion vector field adaptive search technique; PMVFAST, predictive motion vector field adaptive search technique; PSNR, peak signal to noise ratio; EPZS, enhanced predictive zonal search IntroductionDigital video compression is an important technique used to transmit or store digital videos in a limited bandwidth environments or storage areas. H.264/AVC standard 1,2 is one of the modern video coding standards. Compared to past standards, it provides the best balance between the coding efficiency, implementation complexity and cost. Video compression technique depends mainly on the prediction mechanism. Prediction technique in H.264/AVC can be classified into: intra and inter prediction methods. Intra prediction removes spatial redundancies within the frames whereas inter prediction removes temporal redundancies between frames. Inter prediction can be divided into two main processes which is Motion Estimation (ME) and Motion compensation (MC) processes. ME is computationally extensive and consumes about 70% of the total time required for video encoder.The most common ME technique is called Block Matching Algorithm (BMA) where the current frame is divided into Macro blocks (MBs). In BMA each MB in the current frame will be compared with the corresponding MBs in the Search Area (SA) of the reference frame to find the best match MB in the reference frame as shown in Figure 1. Once the best match block for the block in the current frame is located in the reference frame, the motion vector is calculated as the displacement between these two macro blocks. This best candidate will be the motion compensated predictor of the current MB and is subtracted from the current frame, the resulting error is called residual frame. This is called the Motion Compensation (MC) process. The motion vectors and residual frame are adequate information to be sent by H.264 AVC encoder. This information, in general, requires significantly fewer bits than the direct coding of the original frame, and hence improving coding efficiency. The most popular matching criterion in Advanced Video Coder (AVC) is called Sum of Absolute Difference (SAD). The SAD of two N×N MBs is given by equation (1) as follows:Where: is a pixel of the current MB and is a pixel of the reference area. The basic BMA is called Full Search (FS) or Exhaustive Search (ES).Where all the candidates of the SA are considered in the search of the best match predictor. For large SA, FS is computationally extensive and cannot be effective for real time applications, so other suboptimal and fast ME algorithms such as Three Step Search (TSS) AbstractIn real time applications such as video streaming, it is important that the video encoding/decoding is fast. In video compression technique, most of the complexit...