Modelling image completion distortions in texture analysis-synthesis coding

Perception based coding with Texture Analysis and Synthesis (TAS) is a promising way to increase the compression efficiency of modern coding schemes, such as High Efficiency Video Coding (HEVC). TAS approaches typically employ an analysis step which specifies which blocks could be reconstructed by a texture synthesizer. Even though synthesized blocks are perceptually similar to their original versions, they may produce relatively high Mean Squared Error with regard to the original signal. Hence, the contribution of this paper will be a novel image quality assessment method for image completion based on visual attention variations. A subjective experiment has been designed to provide insight into the perceived distortions which are induced by TAS techniques and it is shown that saliency changes are a promising predictor for perceptual distortion. Free access to the image database is provided to encourage further research on this topic.