Efficient non-local means denoising for image sequences with dimensionality reduction

The aim of this paper is to improve both accuracy and computational efficiency of non-local means video (NLMV) denoising algorithm. A technique of principal component analysis (PCA) is used to reduce the heavy dimensionality of patches. A pre-processing step of shot boundary detection is used to split the video sequence into different shots having content-wise similar frames. Further PCA is computed globally for these shots. To speed-up the denoising process, weights are computed in reduced subspace. In the proposed method, we modify the original histogram difference (HD) technique such that content-wise similar frames are separated more systematically and accurately. We have achieved improvement with respect to accuracy and computational speed compared to standard NLM. Moreover, qualitative and quantitative comparisons show that the proposed method is consistently superior compared to that of NLM and some of its variants.