Fast H.264 Mode Selection Using Depth Information for Distributed Game Viewing
Gene Cheung, Antonio J. Ortega, Takashi Okuda Sakamoto · 2008
While H.264 has been known for its superior coding performance over its predecessors, its encoding complexity can be high, especially if an exhaustive search is performed for the rate-distortion optimal mode among many available coding modes for each macroblock. In this paper, we show that in the context of coding 3D graphically rendered frames, depth information—the distance between rendered pixels of objects and the virtual camera popularly used to determine object occlusion during graphics rendering—can be exploited to expedite the mode selection process at a negligible cost of decrease in coding performance. Our proposed techniques take into consideration that regions of similar depth are likely to correspond to regions of uniform motion, an observation that can be exploited to speed-up the selection of block sizes for motion compensation. In our experiments, we show that complexity of the full motion vector search can be reduced by 40 % at the expense of small increase of less than 0.7 % in bitrate and small decrease of less than 0.07dB in PSNR. Moreover, because our proposed schemes operate in the depth domain, we show that they can be easily combined with an existing pixel-domain mode selection algorithm in the literature leading to similar complexity reductions. 1.