A new generic texture synthesis approach for enhanced H.264/MPEG4-AVC video coding

A new generic texture synthesis approach, which is inspired by the work of Kwatra et al., is presented in this paper. The approach is hierarchical, non-parametric, patch-based and for that applicable to a large class of spatio-temporal textures with and without local motion activity. In this work, it is shown, how the new texture synthesizer can be integrated into a content-based video coding framework. That is, the decoder reconstructs textures like water, grass, etc. that are usually very costly to encode. For that, the new texture synthesizer in conjunction with side information that is generated by the encoder is required. The reconstruction of above-mentioned textures at the decoder side basically corresponds to stuffing holes in a video sequence. Spurious edges are thereby avoided by using graph cuts to generate irregular contours at transitions between natural and synthetic textures and preferably place them (the contours) in high-frequency regions, where they are less visible.