A fast noise variance estimation algorithm

Abstract

Noise estimation is an important part of video and image processing. In real time denoising of CMOS sensor environment, it is especially required to achieve fast and accurate noise estimation. This paper presents a fast and efficient noise estimation algorithm. The algorithm is block-based, in which noise of image or video is assumed to be additive zero-mean Gaussian noise. The algorithm requires N1 × N1 window and the corresponding operation of computing variance. In order to achieve fast and accurate noise estimation, the method we use can avoid the sorting process making the complexity order of the algorithm down to O(n) from either O(n <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ) or O(nlog <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> n). In order to improve the accuracy of this algorithm, the main parameters in this paper are set to be adjusted adaptively according to the content of images or videos. By conducting experiments we find that the algorithm is fast and accurate.