Improved Accuracy in Initial Search Center Prediction to Fasten Motion Estimation in h.264/AVC

ABSTRACT

In this paper, a new two-step approach for enhancing the accuracy of initial search center (ISC) prediction in h.264 has been proposed which improves the speed of motion estimation in video encoding. Previous methods for estimating the ISC worked in a single step by finding the mean/median of motion vectors (MVs) of neighboring blocks of the current and reference frames. The major drawback of all the existing ISC techniques is that they consider the participation of all the neighboring blocks with equal probability without taking into account their correlation with the current block. The blocks which have least correlation or no correlation at all affect the accuracy of prediction and hence increase the chances of trapping the search in local minima. Moreover, in the existing ISC prediction techniques, participation of MVs of restricted neighboring blocks is considered, which further limits the prediction accuracy. To elevate these drawbacks, a new two-step approach has been presented. The first step of the method works by identifying some candidate blocks for ISC and the second phase refines the search to obtain best possible ISC. Use of all the surrounding blocks from spatial and temporal frame along with the refinement stage has improved the accuracy. Simulation results clearly show the enhancement in accuracy of ISC prediction, improvement in video quality in terms of peak signal-to-noise ratio (PSNR) and reduction in the number of search steps. Most recent approach in ISC prediction has shown an improvement of 11.7% as compared to standard median predictor, whereas the proposed technique shows an improvement of almost 50%. The reduction in search points is nearly 40%–50% compared to standard median predictor for fast-motion sequences. Also the proposed technique works equally well for fast, medium, slow, cif, qcif and HD video sequences as indicated in the results.

KEYWORDS:

• Initial search center
• Motion vectors
• Motion estimation
• Peak signal-to-noise ratio
• Video coding
• Video compression

Additional information

Notes on contributors

S. Madan Arora

Shaifali Madan Arora is an assistant professor at MSIT, New Delhi. She has completed her B.Tech from GNDU, Amritsar, India and M.Tech from GNDEC, Ludhiana, India. She has more than 12 years of experience in teaching. She is a life member of ISTE. She is a research scholar at GGSIPU, New Delhi. Her research interests include digital image and signal processing, artificial intelligence, microprocessors and controllers. She has various research publications in quality national and international conferences.

E-mail: [email protected]

N. Rajpal

N. Rajpal is professor at USICT, GGSIPU, New Delhi. He did his BSc (Engineering) in Electronics & Communication from R.E.C. Kurukshetra, now known as NIT, Kurukshetra. He did his M.Tech and PhD from Computer Science & Engineering Department, IIT, Delhi. He served in various capacities and has more than 24 years of experience in teaching and research. He has worked as senior scientific officer for more than eight years at Centre for Applied Research in Electronics IIT Delhi on various sponsored and Consultancy projects. Before joining this university in July 2000 as reader, he worked for more than four years as assistant professor at C.R. State College of Engineering, Murthal, where he was In-charge of Computer Science and Engineering Department for about two years. He worked as reader, USICT and In-charge of Computer Center from July 2000 to August 2004 in G.G.S. Indrapratha University. He also worked as Head CS&E at IGIT from January 2005 to December 2007. He has supervised several M.Tech. and three PhD students. He has published/presented more than 75 research papers in National and International Journals/Conferences. He is a life member of CSI and ISTE. His research interests include computer vision, image processing, pattern recognition, artificial neural networks, computer graphics, algorithms design and digital hardware design.

E-mail: [email protected]

K. Khanna

K. Khanna is, at present, an associate professor with North Cap University, Gurgaon, Haryana. She has 15 years of experience in teaching during which she has published more than 20 research papers and guided 15 M.Tech students in their research work. Her research areas include artificial neural networks, digital image processing, computer graphics and design and analysis of algorithms. She has recently submitted her PhD thesis. Apart from that she is working as a Radio Jockey with All India Radio FM Gold.

E-mail: [email protected]

R. Purwar

R. Purwar has obtained his ME (computer science & engineering) degree from MNREC Allahabad (currently known as MNNIT Allahabad). He has pursued his doctorate from university school of information and communication technology (USICT), GGSIP University, Delhi. He is a life member of Computer Society of India (CSI) and Indian Society of Technical Education (ISTE). He has various publications in peer reviewed quality international journals and conferences. His research interests include image/video processing, pattern recognition, video security and database management.

E-mail: [email protected]