A sequential projections based postprocessor for block-based image coding

In this paper, we propose a novel post-processing techinque to reduce blocking artifacts for block-based image coding schemes. Our approach focuses mainly on the reconstruction of the surface continuity, including the continuity of edges and textures in the image objects. To do this, we introduce a set of continuous functions suitable for characterizing edges and, by a linear combination of these functions, estimate the original artifact-free image. We propose the Projections onto the OverComplete Basis (POCB) algorithm to find linear coefficients satisfying a constraint which controls the difference between the blocky and estimated images. The proposed and conventional techniques are tested on various images compressed by the JPEG standard and vector quantization. The simulation results show that the proposed technique yields better results both objectively and subjectively.