Rate distortion bounds for blocking and intra-frame prediction in videos

Recently we revisited the classic problem of developing a correlation model for natural videos by proposing a block-based local-texture-dependent correlation coefficient model for natural videos in the original spatial-temporal domain prior to any processing. We used this conditional correlation model to calculate the conditional rate distortion function when universal side information on local texture is available at both the encoder and the decoder. We demonstrated that this conditional rate distortion bound is indeed a valid lower bound with respect to the operational rate distortion curves of both intra-frame and inter-frame coding in AVC/H.264. Besides the theoretical lower bounds, we also produced a practical lower bound for a single video frame based on the new conditional correlation model and a few assumptions on intra-frame video coding. In this paper we extend the work of generating practical rate distortion bounds for natural videos from a single frame to multiple frames and examine the relationship between texture and motion, the two most dominant visual aspects in natural videos.