Enhanced Intra Prediction Using Context-adaptive Linear Prediction

Intra prediction is a fundamental prediction type in blockbased video coding, which has been adopted by the most recent finalized standard H.264/AVC. H.264/AVC has specified several intra prediction modes that predict the current block through surrounding spatial neighboring pixels in a causal window along various directions. In this paper, we investigate the use of linear prediction for intra block coding and propose a context-adaptive intra prediction approach. Specifically, we use the least square prediction and derive the linear prediction coefficients using reconstructed data. The linear prediction coefficients implicitly embed the local texture characteristics and thus the intra prediction mode is adaptively adjusted according to the local context. No extra overhead is needed for signaling the coefficients since the decoder simply repeats the same deriving process. We take this context-adaptive linear prediction as an additional intra prediction mode along with existing H.264/AVC intra prediction modes, and choose the best mode through rate-distortion optimization. By turning on the new intra prediction mode, we will demonstrate an enhanced coding efficiency performance. The downside of the proposed approach is the increased computational complexity at both the encoder and the decoder.