Denoising 2-dimensional images using the fused lasso signal approximator for improving classification accuracy
Won Son1
1Department of Statistics, Dankook University
Correspondence to: 1 Assistant professor, Department of Statistics, Dankook University, Gyeonggi-do 16890, Korea. E-mail: son.won@dankook.ac.kr
Received May 28, 2023; Revised June 23, 2023; Accepted June 25, 2023.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
- Abstract
- Various sources of noises may contaminate image data. Noise in the image data, not only degrade image quality, but also deteriorate the classification result. In this study, we propose an image denoising procedure based on the two-dimensional fused lasso signal approximator (FLSA) to improve the classification accuracy for noisy image data. The FLSA gives a piecewise constant mean structure for each interval using the least squares method with total variation penalty. In the case of the two-dimensional FLSA, the total variation penalty yields block structure in which adjacent cells in a block have the same mean value. Therefore, if we apply the two-dimensional FLSA for denoising image data, then we can expect that the noise intensity will be weakened by integrating the noisy cells into a block with the adjacent cells. Furthermore, the L1 penalty in the two-dimensional FLSA can force the mean levels of noise corrupted blocks to zeros. Applying the proposed procedure to the MNIST data, we found that the classification accuracy of noisy image data can be improved.
- Keywords : Classification, fused LASSO signal approximator, image data denoising, total variation denoising
Full Text(PDF) Free
