Noise Level Estimation for Image Processing

Abstract

This paper compares the results between two well known noise level estimation algorithms. We address the issue of estimating the variance of additive white Gaussian noise in digital images, even with large textured areas. Noise can significantly influence the quality of digital images. This method is based on two approaches. Both algorithms were introduced and we tested both algorithms with 5 artificial images and 8 natural images. One approach searches intensity-homogeneous blocks and then we estimate the noise variance in these blocks. The other method is based on wavelet transform, we obtain variance of additive white Gaussian noise using coefficients of wavelet transform. Based on the results, we adaptively choose the method and obtain the most appropriate noise level.