This paper describes a study aimed at comparing the real image sensor noise distribution to the models of noise often assumed in image denoising designs. Quantile analysis in pixel, wavelet, and variance stabilization domains reveal that the tails of Poisson, signal-dependent Gaussian, and Poisson-Gaussian models are too short to capture real sensor noise behavior. Noise model mismatch would likely result in image denoising that undersmoothes real sensor data.

Noise is present in all images captured by real-world image sensors. Poisson distribution is said to model the stochastic nature of the photon arrival process and agrees with the distribution of measured pixel values. We propose a method for estimating unknown noise parameters from Poisson corrupted images using properties of variance stabilization. With a significantly lower computational complexity and improved stability, the proposed estimation technique yields noise parameters that are comparable in accuracy to the state-of-art methods.