Noise level estimation based on eigenvalue learning

At present, many algorithms use a single minimum eigenvalue to estimate the real noise level, and the levels estimated by these algorithms have been proven to be less than the real noise levels, this is known as underestimation. To address this problem, this paper uses multiple eigenvalues to obtain the relationship between eigenvalues and the real noise level through sample training, calculates the learning coefficients for different noise levels in the relationship expression by linear fitting, and then inputs the learning coefficients into the noise image for noise level estimation. Experiments demonstrate that the algorithm proposed in this paper can significantly improve the underestimation problem of the traditional algorithm and has better estimation accuracy for various noise levels in gray images, color images, and texture images of various scenes.

Automated summary

This paper addresses the problem of underestimation of noise levels in traditional algorithms by using multiple eigenvalues to estimate the real noise levels. Experimental results demonstrate that the proposed algorithm has better estimation accuracy for various noise levels in different image scenes.