Title:
SYSTEM AND METHOD FOR TRAINING OF NOISE MODEL USING NOISY SIGNAL PAIRS
Document Type and Number:
WIPO Patent Application WO/2023/085899
Kind Code:
A1
Abstract:
A noise model is iteratively trained to simulate introduction of noise by a capture device, by use of a denoiser and a training data set of pairs of noisy signals. First and second noisy signals of each pair are independently sampled by the capture device from source information corresponding to the pair. Each iteration of training obtains first and second denoised signals from respective noisy signals, then optimizes at least one loss function which sums first and second terms to train both the noise model and the denoiser, where the first term is based on the first denoised signal and the second noisy signal, and the second term is based on the second denoised signal and the first noisy signal. By using noisy samples, the complexities of obtaining "clean" signals are avoided. By using "cross-sample" loss functions, convergence on undesired training results is avoided without complex regularization.
More Like This:
Inventors:
MALEKY ALI (US)
BRUBAKER MARCUS ANTHONY (US)
BROWN MICHAEL SCOTT (US)
BRUBAKER MARCUS ANTHONY (US)
BROWN MICHAEL SCOTT (US)
Application Number:
PCT/KR2022/017960
Publication Date:
May 19, 2023
Filing Date:
November 15, 2022
Export Citation:
Assignee:
SAMSUNG ELECTRONICS CO LTD (KR)
International Classes:
G06T5/00; G06N3/08
Foreign References:
| US20200234080A1 | 2020-07-23 | |||
| US9779727B2 | 2017-10-03 | |||
| KR20210116923A | 2021-09-28 | |||
| US20190378247A1 | 2019-12-12 | |||
| US20170084286A1 | 2017-03-23 |
Attorney, Agent or Firm:
KIM, Tae-hun et al. (KR)
Download PDF: