Image noise level estimation based on affine reconstruction and noise sample histogram

An image noise level estimation method was presented by using affine reconstruction technique and the calculated noise sample histogram. The watershed-based image segmentation was firstly utilized to divide the noisy image into several homogenous blocks. Then by applying affine reconstruction technique, the noiseless affine image signal and the noise residual image were obtained. Noise samples for the standard deviation values of each segmented patch were calculated from the noise residual image. After that the histogram of estimated noise samples was described to find out the specific noise level interval with the most noise samples falling into. Finally, the image noise standard deviation was computed by the average of noise samples in the selected noise interval. Experiments are implemented to demonstrate the effectiveness of the proposed algorithm. The presented method could produce accurate and reliable estimation results for images with rich textures and edges.