A Content Aware Fast Motion Estimation Algorithm for H.264/AVC Humaira Nisar, and Tae-Sun Choi, Senior Member, IEEE Gwangju Institute of Science and Technology, Korea (humaira, tschoi)@gist.ac.kr Abstract In this paper we propose a content aware fast block matching motion estimation algorithm that can reduce the computation complexity of motion estimation module in H.264/AVC while maintaining almost same coding efficiency. In the first step inter block correlation is used to define a correlation co efficient that helps to identify the motion content of the current block. Based on this co efficient the algorithm adopts different search strategies for different video contents to get more efficient matches. Implementation of the proposed algorithm in JM12.2 of H.264/AVC shows reduction in encoding time while maintaining almost same bit rate and PSNR. 1. Introduction Block based motion estimation (ME) and compensation is an essential part of several video coding standards such as MPEG-1/2/4 [1,2], ITU-T H.261 [3] and H.264/AVC [4], that helps to reduce the temporal redundancy within frames and leads to higher compression ratio. Motion estimation is usually the most computationally intensive part in a typical video encoder (about 50%-90% of the entire system), therefore an efficient ME module is essential and vital. Full Search (FS) is the most straightforward and optimal block matching algorithm (BMA) that exhaustively searches inside the search window for the best match block hence it is computationally expensive. Several fast ME algorithms have been proposed so far. Diamond Search (DS) is an outstanding algorithm adopted by MPEG-4 verification model (VM) [5] due to its superiority to other methods in the class of fixed search pattern algorithms. However fixed search patterns are unable to constantly match the dynamic motion content, thus not only wasting the computational power but also leading to local minimum matching error trapping and large prediction errors. UMHexagonS [6] is a novel motion estimation algorithm that has been adopted by H.264/AVC standard as one of the fast motion estimation algorithms. 2. Proposed algorithm Many objects in a scene are composed of more than one block so the motion field of a block can be tracked from neighboring blocks both in spatial and temporal domains. This correlation information has been utilized by many ME algorithms to improve the accuracy of predicted motion vector (MV). In this paper we will combine spatio-temporal correlation information and the knowledge of the motion content of the video to speed up the motion estimation process while maintaining coding efficiency close to FS algorithm. 2.1. Motion Vector Prediction The spatio temporal neighborhood employed in our algorithm consists of two neighboring blocks from the current frame (left and above block) and the collocated block (the block at the same location in the reference frame), as shown in Fig. 1. It is observed in real world video sequences that the median spatial scheme is suitable for sequences having uniform motion whereas the motion vector of the collocated block gives better results for sequences having complex motion when the neighboring blocks belong to different objects or have different motions. Fig. 1. Blocks for spatio-temporal correlation information Current Block ROS Reference Frame (n-1) R Current Frame n B A 2008 Second International Conference on Future Generation Communication and Networking Symposia 978-0-7695-3546-3/08 $25.00 © 2008 IEEE DOI 10.1109/FGCNS.2008.34 214 Authorized licensed use limited to: Kwangju Institute of Science and Technology. Downloaded on June 23, 2009 at 03:20 from IEEE Xplore. Restrictions apply.