Gaussian noise parameter estimation based on multiple singular value decomposition and non‐linear fitting
Jinli Qi, Lei Sun, Kengpeng Li, Lingang Wang · IET Image Processing · 2022
Abstract Noise standard deviation (STD) is an important parameter in many digital image processing applications. This paper presents a Gaussian noise parameter estimation algorithm using multiple singular value decomposition (SVD) and non‐linear fitting. The proposed algorithm adds known noise to the original noise image many times to generate a noise‐corrupted image set and then performs SVD on each image. By analyzing the singular values of the noise‐corrupted images, an overdetermined equation system with respect to the noise STD is established. The Gauss–Newton iteration method and backtracking Armijo line search are used to solve the equations, which improve the convergence speed and reduce computational cost. Compared with other methods, the mean error of the proposed algorithm on the TID2008 dataset is 0.028, which is several times lower than other methods. This shows that the performance of our estimator is significantly improved.