Adaptive non-local means filtering for image deblocking

In this paper, a novel deblocking algorithm is proposed to remove the blocking artifacts in images compressed by block discrete cosine transform (BDCT), using an adaptive non-local means filter. We prefer block as the basic processing unit, whose estimation is obtained as a weighted sum of neighborhood blocks. The weights are defined as a multivariate Gaussian with a covariance matrix that depends on the difference between the quantization noise of the block and that of its neighborhood blocks. The weights react adaptively to both block content and quantization noise intensity. Extensive experimental results and comparative studies are proposed to demonstrate the effectiveness of the proposed deblocking algorithm.