Intra predictive transform coding based on predictive graph transform
Yongzhe Wang, Antonio Ortega, Gene Cheung · 2013
In this paper, we propose a new intra-frame coding approach using the predictive graph transform (PGT). The predicted block together with the reference pixels are modeled as a normal distributed random vector with respect to a graph whose edges represent the correlations between pixels. This model is more flexible than the Gaussian Markov random field (GMRF) model in the sense that it enables us to adapt the graph both before and after the collection of the statistics. The optimal prediction and the transform of the prediction residual are then derived jointly. Two PGT based intra coding schemes are proposed: one is based on global image statistics and the other is mode-adaptive, i.e., the graph is adaptive to different directional modes defined in H.264/AVC. The simulations show the advantage of our proposed approach over standard intra predictive transform coding in terms of both prediction quality and coding gain assuming the model parameters are known at decoder.