Graphics-to-video encoding for 3G mobile game viewer multicast using depth values

Towards the goal of streaming mobile network game sequences in standard compliant video to thousands of game observers, a specialized video encoder called grencoder that takes 3D game graphics as input is discussed. In particular, we present a framework that increases the visual quality of regions of interest (ROI) of the video by performing intelligent mode selection during h.263+ video encoding. The regions of interest are identified by extracting the depth values of pixels in the frame buffer made available when 3D objects are projected to a 2D surface during rasterization. Results show that by performing ROI mode selection, PSNR of test sequence slightly increases while subjective quality of nearby objects increases noticeably.