Reduction of blocking artifacts by adaptive DCT coefficient estimation in block-based video coding

We propose an adaptive post-processing method for reducing blocking artifacts in block-based video coding. This method is based on DCT coefficient estimation by quadratic programming method for each block. For the blocks including detected edges, we define an objective function on the basis of preservation of the edges and continuity of the pixel values at the block boundaries. For the other blocks, we define an objective function on the basis of the continuity around the block boundaries. Furthermore, the DCT coefficients are estimated sequentially from lower frequency side to higher in order to obtain the stable estimates. For the blocks coded with the use of motion-compensated prediction, we estimate the DCT coefficients representing the prediction errors. The experiments show that the proposed method improves both subjective and objective image quality compared to the H.263 decoded frames and also improves the SNR of the sequences coded by the H.263+ Annex.J method at several bit rates.