Video noise reduction using noise bias compensation

抄録

Image noise amplified by brightness adjustment contains two components: the noise bias (NB; where the mean of the noise has a nonzero value) and the noise variance. Current conventional filters deal with the noise variance by using an approach that focuses on the correlation between pixels; however, these filters do not deal with the NB. The noise bias compensation (NBC) method is a new noise reduction approach that focuses on the NB. NBC reduces the image noise by subtracting the image correction from the observed pixel values and thus is faster than the currently existing filters. However, NBC is a noise reduction method for use with still images (2D images) and must therefore be extended for use with video sequences (3D images). In this paper, we propose an approach to extend the NBC method to video processing by selecting a compensation value from several pre-prepared values based on the frame features. These pre-prepared compensation values are modeled using all frames from the video without noise and before brightness adjustment.