A nonlocal means based adaptive denoising framework for mixed image noise removal

Most existing image denoising algorithms can only deal with a single type of noise; however, real world images are typically contaminated by more than one type of noise during image acquisition and transmission process. Recently, nonlocal approaches got great success in removing Gaussian noise; however, they cannot deal with impulse noise due to their nature. In this paper, we propose an improved nonlocal means (NL-means) to simultaneously remove impulse noise and Gaussian noise. An adaptive mixed noise removal framework based on the improved NL-means is also introduced. Comparing with existing NL-means based mixed noise removal frameworks which remove one type of noise at a time; the proposed framework can remove mixed noise simultaneously in a single step. Experimental results show that the proposed denoising framework has better denoising performance for mixed noise; meanwhile it is much more efficient which makes future parallel optimization such as GPU optimization possible.