Laser & Optoelectronics Progress, Volume. 58, Issue 6, 610018(2021)
NVST Image Denoising Based on Self-Supervised Deep Learning
Fig. 1. Blind spot convolution network with four branches
Fig. 2. Convolution blind spot network after optimizing branch
Fig. 3. Blind spot receptive domain. (a)Upper half plane;(b) left half plane; (c)lower half plane;(d) right half plane; (e)complete receptive region with blind spots after combination
Fig. 4. Denoising results of each method. (a) Original images; (b) ground truth; (c) noise interference; (d)proposed algorithm; (e) CBSN; (f) Noise2Void; (g) BM3D; (h) Gaussian filter; (i) mean filter
Fig. 5. Correlation diagrams. (a) Proposed algorithm;(b)CBSN; (c)Noise2Void; (d)BM3D; (e)Gaussian filter; (f)mean filter
Fig. 6. Power spectra
Fig. 7. Denoising results of real sun image. (a) Real images; (b)proposed algorithm; (c) CBSN; (d) Noise2Void; (e) BM3D; (f) Gaussian filter; (g) mean filter
|
Get Citation
Copy Citation Text
Lu Xianwei, Liu Hui, Shang Zhenhong. NVST Image Denoising Based on Self-Supervised Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(6): 610018
Category: Image Processing
Received: Aug. 1, 2020
Accepted: --
Published Online: Mar. 16, 2021
The Author Email: Zhenhong Shang (shangzhenhong@126.com)