Fast Non-Local algorithm for image denoising

摘要

In this paper, improvements to non-local means (NLM) image denoising method is proposed to reduce the computational complexity. In the original NLM algorithm, neighborhood weightages are computed using the window similarity technique. The proposed technique replaces the window similarity by a modified multi-resolution based approach with much fewer comparisons rather than all pixels comparison. This approach also uses the concept of filtering out non-similar neighborhood pixels based on fixed sized window gray mean values. Further, mean values of the variable sized windows in the image are computed efficiently using summed image (SI) concept, which requires only 3 additions. The proposed approach is nearly 80 times faster than original Baudes NLM algorithm with close subjective and objective quality measurements.