A novel image deblocking approach within a graph framework

Abstract Since JPEG compression is achieved via independent and coarse quantization of discrete cosine transformation coefficients in each block, its results generally exhibit visually annoying blocking artifacts and Gibbs phenomena at low bitrates. Moreover, the existing image deblocking algorithms are basically processed under the classical image topology structure. In this paper, we present a novel image blocking artifact-free methodology within a graph framework. A graph is a compatible tool. Most of the existing image deblocking models can be improved by transforming images or image patches into graph signals. To verify this viewpoint, we focus on improving the existing mainstream total-variation-based and constrained non-convex low-rank image deblocking models. Furthermore, we provide numerical algorithms and the corresponding fast workarounds. The introduction of a graph renders the pixel representation more flexible and adaptive, and can improve the computational efficiency of the constrained non-convex low-rank image deblocking approach. In addition, twelve standard images are used to test the performance of the proposed models. The experiments show that in comparison to several existing famous models, our models achieve competitive results in terms of deblocking and calculation speed.