Skip to Main Content
This paper introduces an adaptive postprocessing method in block-based discrete cosine transform (BDCT) coded images. We present a framework, which is a serial-concatenation of a 1-D simple deblocking filter and a 2-D directional filter. First, we classify the simple deblocking filter by an adaptive threshold depending on local statistical properties, and update block types appropriately by a simple rule, which affects the performance of deblocking progresses. Second, a 2-D directional filter is introduced to suppress the ringing artifacts caused at the sharp edges and the block discontinuities while preserving true edges and textural information. Finally, comprehensive experiments indicate that the proposed algorithm outperforms lots of deblocking methods in the literature, in terms of evaluation of PSNR and subjective visual qualities.