Engineering Research Center of Network Management Technology for High Speed Railway Ministry of Education, School of Computer and Information Technology, Beijing Jiaotong University, No. 3 Shangyuan Village, Haidian District, Beijing, China
Noise level is an important parameter in many visual applications, especially in image denoising. How to accurately estimate the noise level from a noisy image is a challenging problem. However, for color image denoising, it is not that the more accurate the noise level is, the better the denoising performance is, but that the noise level higher than the true noise can achieve a better denoising result. For better denoising, we propose a statistical iterative method based on low-rank image patches. We select the low-rank patches in the image and calculate the eigenvalues of the covariance matrix of these patches. Unlike the existing methods that take the smallest eigenvalue as the estimated noise level, the proposed method analyzes the relationship between the median value and the mean value of the eigenvalue according to the statistical property and selects an appropriate number of eigenvalues to average as the estimated noise level. Extensive experiments are conducted, demonstrating that our estimated noise level reaches the highest value of all comparison methods. And the denoising results on color images of our method outperform all the state-of-the-art methods and the true noise level.
Varun Mannam, Yide Zhang, Yinhao Zhu, Evan Nichols, Qingfei Wang, Vignesh Sundaresan, Siyuan Zhang, Cody Smith, Paul W. Bohn, and Scott S. Howard Optica 9(4) 335-345 (2022)
44. S. Roth and M. J. Black, “Fields of experts,” Int. J. Comput. Vis. 82, 205–229 (2009). [CrossRef]
45. D. Martin, C. Fowlkes, D. Tal, and J. Malik, “A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics,” in IEEE International Conference on Computer Vision (ICCV) (IEEE Comput. Soc., 2001).
Cited By
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
The noise values in the first column are the noise levels added to the R, G, and B channels, respectively. The next five columns are the average of the noise levels of R, G, and B channels estimated by each method.
Table 3.
Accuracy Results of Compared Methods for Data Set Kodak
The noise values in the first column are the noise levels added to the R, G, and B channels, respectively. The next five columns are the average of the noise levels of R, G, and B channels estimated by each method.
Table 6.
Accuracy Results of Compared Methods for Data Set BSD500
The noise values in the first column are the noise levels added to the gray-scale images. The next five columns are the average noise levels estimated by each method.
Table 9.
Accuracy Results of Compared Methods for Gray-Scale Image Data Set BSD68
The noise values in the first column are the noise levels added to the R, G, and B channels, respectively. The next five columns are the average of the noise levels of R, G, and B channels estimated by each method.
Table 3.
Accuracy Results of Compared Methods for Data Set Kodak
The noise values in the first column are the noise levels added to the R, G, and B channels, respectively. The next five columns are the average of the noise levels of R, G, and B channels estimated by each method.
Table 6.
Accuracy Results of Compared Methods for Data Set BSD500
The noise values in the first column are the noise levels added to the gray-scale images. The next five columns are the average noise levels estimated by each method.
Table 9.
Accuracy Results of Compared Methods for Gray-Scale Image Data Set BSD68