Noise variance estimation based on image segmentation

A new two-step noise variance estimation algorithm was proposed based on image segmentation. In the first step, a noisy image was smoothed and was segmented by the statistical region merge (SRM) algorithm, then the variance of each region was computed, and some regions were selected based on the statistical rule to estimate the noise variance. In the second step, the parame-ters of filtering, segmentation and estimation were revised according to the estimated noise variance, and a new cycle of image filte-ring, segmentation and estimation was performed to obtain more accurate estimation. Experimental results on large numbers of images and various noises show that the proposed algorithm can estimate the noise variance quickly and accurately.