Characterization and adaptive texture synthesis-based compression scheme

This paper presents an adaptive texture synthesis-based compression scheme, where textured regions are detected and removed at encoder side, allowing the decoder to use texture synthesis to fill them. The detection relies on locally adaptive resolution segmentation. According to results shown by synthesis algorithms, they need to be parameterized according to the patterns to be synthesized. In this framework, the synthesizer gets its parameters from DCT feature-based texture descriptors. An adaptive pixel-based algorithm is used, relying on the comparison between current pixel neighborhood and those in an atypically shaped sample. Different neighborhood sizes are considered to better catch texture patterns. The framework has been validated within an H.264/AVC video codec. Experimental results show significant bit-rate saving at similar visual quality.