Image noise estimation method based on sparse principal component analysis

Realizing accurate estimation of noise parameters in digital images is of great significance for improving the quality of image processing. When the sparse principal component analysis is performed on the image contaminated by Gaussian white noise, the mean value of the load vector of some principal components and the standard deviation of Gaussian white noise show a certain linear relationship. Based on this feature, this paper proposes a fast and accurate image noise estimation method. In this method, a plurality of new Gaussian white noises of known standard deviation levels are added to the Gaussian white noise-contaminated image to generate a plurality of new images, and then each sample is subjected to sparse principal component analysis and mean of multiple principal component load vectors are obtained. Finally, an accurate estimation of the standard deviation of Gaussian noise is achieved by solving an over determined system of equations. The experiment results show that the method has high estimation accuracy under low noise (delta(0) = 5) to high noise (delta(0) = 70) conditions, and strong robustness is demonstrated. This method has certain practical value in practical engineering.