Teleconferencing and other applications of 3D video have become of increasing interest. Besides stereo display, realistic 3D requires also free viewpoint video (FVV), in order to provide adequate motion parallax. One of the most efficient ways of synthesizing FVV of natural scenes is by image based rendering, based on multiview video. A new approach for multiview coding has been recently proposed, and shows that encoding each macroblock of each frame with appropriate quantization parameters (QPs) improves compression by around 2X. The QPs used to acheive such improvement are tuned according to an adhoc relation to the weights used in the synthetic view. In this paper we show that the ad-hoc relation is actually optimum for a Guassian source, and derive the optimum QPs for the generic case.