Patch-adaptive non-local means filtering for image deblocking using a no-reference metric

Annoying blocking artifacts are exhibited in images compressed by block discrete cosine transform (BDCT), primarily at low bit-rate. Many algorithms have been proposed to alleviate this degradation by removing quantization noise, under the condition that the noise magnitude is known. However, in many applications the receiver only gets the degraded image without any prior knowledge about noise distributions, in which case the usage of the existing deblocking algorithms are significantly limited. In this paper, we propose a novel patch-adaptive non-local means filter for deblocking using no side information about noise distribution. To do this work, a no-reference metric is first given to evaluate the deblocking efficiency. Then, the non-local means filter is applied block by block and the metric is used to choose the optimal filtering degree adaptively. The result of proposed deblocking algorithm is superior to conventional approaches in both subjective and objective aspects, which are supported by extensive experiments.