Image SNR estimation using the autoregressive modeling

A number of techniques have been proposed during the last two decades for Signal-to-Noise Ratio (SNR) estimation in images. The majority of these techniques are based on the cross-correlation function of two images of the same area. However, the need for two images to estimate SNR value confines these techniques to non-stored images and thus limits their applications. In this paper the second order statistics of image corrupted by additive white noise are modeled by Autoregressive-model and the relationship between AR model and linear prediction is utilized in estimating the predictor coefficients. The predictor is then used to estimate the zero-offset autocorrelation value and accordingly obtain the power of the noise-free image. Unlike others, the proposed technique is based on single image and offers the required accuracy and robustness in estimating the SNR values.