Single-image signal-to-noise ratio estimation

A method for estimating the signal-to-noise ratio from a single image is pr esented in this paper. The autocorrelation-bused technique requires that im age details be correlated over distances of a few pixels, while the noise i s assumed to be uncorrelated from pixel to pixel. The latter is shown to be a good approximation in the case of scanning electron microscope (SEM) ima ges provided that the video signal is not band limited. The noise component is derived from the difference between the image autocorrelation at zero o ffset and an estimate of the corresponding noise-free autocorrelation. Nonl inear effects introduced by intensity saturation and their implications on the image signal-to-noise ratio are also discussed.