Accurate Noise Level Estimation through Singular Values and Linear Regression

Accurate estimation of noise level in digital images is of critical importance prior to application of denoising techniques in the areas of image processing and computer vision. In this paper, we propose a novel approach for estimation of noise level using: (i) Singular values of noisy image from rear end of Singular Value Decomposition (SVD) subspaces (in order to restrict the influence of image content), and (ii) Use of linear regression technique for determination of content dependent parameter which widens the application scope of the proposed method. The experimental results substantiate the robust behavior of the proposed approach over a wide range of image content and reliable estimation of noise level.