A Robust and Fast Non-Local Means Algorithm for Image Denoising

In the paper, we propose a robust and fast image denoising method. Theapproach integrates both Non-Local means algorithm and LaplacianPyramid. Given an image to be denoised, we first decompose it intoLaplacian pyramid. Exploiting the redundancy property of Laplacianpyramid, we then perform non-local means on every level image ofLaplacian pyramid. Essentially, we use the similarity of imagefeatures in Laplacian pyramid to act as weight to denoise image. Sincethe features extracted in Laplacian pyramid are localized in spatialposition and scale, they are much more able to describe image, andcomputing the similarity between them is more reasonable and morerobust. Also, based on the efficient Summed Square Image (SSI) schemeand Fast Fourier Transform (FFT), we present an accelerating algorithmto break the bottleneck of non-local means algorithm --- similaritycomputation of compare windows. After speedup, our algorithm is fiftytimes faster than original non-local means algorithm. Experimentsdemonstrated the effectiveness of our algorithm.