Noise Level Estimation in Images Using Haar Wavelets

Abstract

The paper investigates the possibility and expediency of using the Haar wavelet transform in the problem of estimating the level of discrete Gaussian noise in an image. An algorithm is proposed that uses Haar wavelets to obtain an estimate of the variance of discrete Gaussian noise in a digital image. To reduce the influence of image fragments with a large proportion of high-frequency oscillations of the useful signal, the image is divided into blocks, followed by the selection of blocks with a minimum dispersion. The proposed algorithm is compared with a method based on the use of difference operators for estimating the noise level. This method gives fairly accurate noise variance estimates and has low computational complexity. The results of estimating the variance of the noise of different intensity superimposed on the image by compared methods are presented. Based on the theoretical provisions and the results of experimental studies, it is concluded that the proposed algorithm has the best accuracy in estimating the noise level at lower computational costs.