Segmentation methods for noise level estimation and adaptive denoising from a single image

In order to significantly remove noise, many computer vision algorithms require that the image noise level is known. In this paper, we present two simple but effective methods to automatically estimate the noise level from a single image using improved adaptive graph-based segmentation. The first method can estimate the noise standard deviation of image contaminated by Gaussian noise or natural noise. The second method is based on point-wise estimation of the image noise. Both of them can be applied to attenuate real noise produced by CCD camera, even if the noise is inhomogeneous. The main advantage of our algorithms compared with others is its speed: its complexity is a linear function of the number of the image pixels. This speed allows noise level estimation to be applied in real-time system. At last, we illustrate the utility of the noise estimation for constant time bilateral filter. Meanwhile, experimental results demonstrate the effectiveness of our algorithms.