Fast log-Gabor-based nonlocal means image denoising methods

This paper explores the possibility of incorporating log-Gabor features into nonlocal means image denoising framework. It is found that log-Gabor features are better choice for this task than previously studied geometrical features. Moreover, we combine log-Gabor features with original image patch information to arrive at mixed similarity measure, which leads to further denoising performance improvement. In addition, we test a random projection-based approach to nonlocal means speed-up, guided by the well-known Johnson-Lindenstrauss lemma. Experimental results are quite encouraging.