Motion vector quantization for efficient low bit-rate video coding M. Cagnazzo a , M. A. Agostini b , M. Antonini b ,G. Laroche c , J. Jung c a D´ epartement TSI , TELECOM ParisTech. 37-39 rue Dareau, 75014 Paris (France); b Laboratoire I3S, UNSA/CNRS. 2000 route des Lucioles, 06903 Sophia Antipolis Cedex (France); c France T´ el´ ecom R&D, MAPS/MDG/SIA. 38-40 rue Leclerc, 92794 Issy Les Moulineaux (France) ABSTRACT The most recent video coding standard H.264 achieves excellent compression performances at many different bit-rates. However, it has been noted that, at very high compression ratios, a large part of the available coding resources is only used to code motion vectors. This can lead to a suboptimal coding performance. This paper introduces a new coding mode for a H.264-based video coder, using quantized motion vector (QMV) to improve the management of the resource allocation between motion information and transform coefficients. Several problems have to be faced with in order to get an efficient implementation of QMV techniques, yet encouraging results are reported in preliminary tests, allowing to improve the performances of H.264 at low bit-rates over several sequences. Keywords: Video coding, motion vector, quantization, H.264 1. INTRODUCTION An efficient resource allocation between motion vectors (MV) and motion-compensated residual is a key feature in any video coder aiming at good rate-distortion (RD) performances. In standard coders 1, 2 it is only possible to indirectly choose how the bit-rate is shared between motion and residual by selecting one among the several available coding modes for each macroblock (MB). As a consequence, it has been noted that when a sequence is encoded at low and very low bit-rates, a large quota of resources is allocated to MVs. 3 This suggests that, in the framework of a H.264-like coder, there could be room for performance improvement if some new coding mode with less costly motion information is introduced. This intuition is reinforced by some simple quantitative study. In Fig. 1 we report the average macroblock rate and distortion for several coding modes in a H.264 coder (JM v.11.0 KTA 1.4, Ref. 4 ). These operation points have been obtained on the sequence city ; for other sequences similar results have been obtained. We see that there is a significant gap between the low-cost, high-distortion SKIP mode and the relatively higher-cost, low-distortion INTER 16x16 mode (while INTRA and lossless IPCM modes are far more expensive and usually not suitable in low bit-rate context). Therefore, we want to introduce a new coding mode which should have an intermediate behavior between SKIP and INTER 16x16. On one hand, the introduction of new coding modes increases the signalling cost (i.e. the coding cost of any selected mode), but on the other, it hopefully has a better RD performance for some MB. The main target of this paper is to verify that gains associated to the new mode surpass the losses. To achieve this target we propose the lossy coding * of motion vectors, obtained via quantization. Moreover this lossy coding is performed in an open loop system so that, while the motion-compensated residual is computed with the motion vector v, the vector v itself is quantized to ˜ v before being sent to the decoder. This will reduce the coding rate, but can also increase the distortion as the quantized vector ˜ v will be used to compute the motion-compensated (MCed) prediction Further author information: E-mail: cagnazzo@telecom-paristech.fr, {agostini,am}@i3s.unice.fr {guillaume.laroche,joelb.jung}@orange-ftgroup.com * It is worth noting that in Ref. 3 authors achieved significant rate reduction by lossless coding of MVs.