Surface fitting approach for reducing blocking artifacts in low bitrate DCT decoded images

The global nature of the discrete cosine transform and the quantisation of its coefficients are responsible for the blocking artifacts in low bit-rate block-based decompressed JPEG images. At bit-rates lower than 0.5 bpp, these artifacts become visibly annoying, particularly in flat and low activity areas of the image. Most deblocking techniques to date are block-based and have only limited visual success because of the highly regular grid structure of the block boundaries. This paper is concerned mainly with the reduction of blocking artifacts in these smooth areas using a ternary polynomial surface to model the pixel intensity over a 2×2 block array called macro-block. Unlike current polynomial surface fitting techniques, including the JPEG technique, our proposed technique attempts to compensate for the quantisation error in the JPEG compressed DCT coefficients. The technique uses a linear programming approach to optimise the fitting of the model to the decompressed (quantised) DCT coefficients to within their corresponding JPEG quantisation range