Homogeneity classification for signal-dependent noise estimation in images
Meisam Rakhshanfar, Aishy Amer · 2014
This paper presents a fast method to estimate the noise level in real images, and attempts to solve clipping and signal-dependency problems for robust noise estimation. We propose an intensity-variance homogeneity classification technique to classify images corrupted with additive Poisson-Gaussian noise based on intensity and variance. Benefiting from signal-independency in each intensity class, this method localizes the noise-representative homogenous regions in the image. Experimental results show the proposed method rivals state-of-the-art estimation approaches, while it is fast.