Model based multi-pass macroblock-level rate control for visually improved video coding

Visual quality of video depends not only on global measures (like average PSNR over a sequence) but also on how the error is distributed in space and time. Typical macroblock layer rate control schemes adapt the quantization parameters (QP) over a frame to meet the frame level bit target. However, large variations of QP over a coherent spatio-temporal region results in visually displeasing artifacts. Thus, it is important to maintain smoothly varying QP for each visually significant region. In this paper we present a multipass scheme that achieves such a QP adaptation that is also commensurate with the visual significance of each region in the scene. Such a true multipass scheme is then approximated by a rate-distortion (R-D) model-based pseudo-multipass scheme that is suitable for real-time implementation.