Adaptive non-local means filter for image deblocking

Blocking artifacts often exist in the images compressed by standards, such as JPEG and MPEG, which causes serious image degradation. Many algorithms have been proposed in the last decade to alleviate this degradation by reducing the quantization noise. Unfortunately, these algorithms only produce satisfying results under an unreasonable assumption that noise magnitude has been given. However, in most applications, the user only gets inferior image copy, without any side information about noise distribution, therefore the efficiency of existing denoise algorithms is significantly reduced. In this paper, a new metric is first given to evaluate the blocking artifacts; and then non-local means filter is applied to remove quantization noise on the blocks. During the process, nonlocal means filters with different variances are used to do deblocking, and their efficiencies are recorded as the references. The deblocked image is finally the one combined with all blocks filtered with the optimal parameters. We prove with experimental results that the proposed algorithm constantly outperforms the peer ones on all kinds of images.