Laser & Optoelectronics Progress, Volume. 62, Issue 6, 0637004(2025)
Interpretable Deep Learning Image Restoration Algorithm with L2-Norm Prior
Figures & Tables(7)
Fig. 2. Network structure of the interpretable deep learning image restoration algorithm with L2-norm prior
Fig. 3. Comparison of different image restoration methods on the UC Merced dataset
Fig. 4. Comparison of different image restoration methods on real remote sensing data. (a) (b) Remote sensing images by the Landsat8 satellite; (c) (d) remote sensing images by the Zhuhai-1 satellite
Fig. 5. Iterative process of image restoration by proposed method on the UC Merced data
Table 1. Comparison of different image restoration methods on various scene classes of UC Merced
View tableTable 1. Comparison of different image restoration methods on various scene classes of UC Merced
Feature type PSNR /dB SSIM RSA SRN-DeblurNet MIMO-UNet Uformer MISCFilter Proposed RSA SRN-DeblurNet MIMO-UNet Uformer MISCFilter Proposed Average 25.66 29.66 31.56 31.57 31.76 33.34 0.8149 0.8892 0.9094 0.8473 0.9132 0.9137 Agricultural 24.37 25.22 26.02 26.63 25.84 26.76 0.7337 0.7023 0.7300 0.6560 0.7111 0.6922 Airplane 25.41 30.80 35.20 34.97 35.03 35.90 0.8751 0.9477 0.9700 0.9058 0.9655 0.9663 Baseballdiamond 31.96 36.30 38.52 38.52 38.33 40.38 0.9202 0.9625 0.9731 0.9252 0.9708 0.9828 Beach 28.15 33.31 33.80 34.51 33.07 36.43 0.8439 0.9201 0.9258 0.8881 0.9195 0.9400 Buildings 21.91 30.02 33.25 32.78 34.38 36.09 0.7420 0.9061 0.9448 0.8880 0.9537 0.9783 Chaparral 25.83 29.50 29.50 30.50 31.28 31.62 0.7534 0.9030 0.8952 0.8563 0.9242 0.9325 Denseresidential 20.40 28.31 33.06 33.28 34.20 34.70 0.7473 0.9202 0.9681 0.9350 0.9754 0.9630 Forest 32.40 31.25 33.62 33.28 32.68 34.18 0.8943 0.8823 0.9035 0.7982 0.9002 0.9654 Freeway 21.15 33.52 37.02 35.28 37.05 38.02 0.7470 0.9543 0.9758 0.9635 0.9453 0.9802 Golfcourse 32.81 33.83 34.66 34.74 34.92 36.58 0.9256 0.9315 0.9349 0.8682 0.9394 0.9475 Harbor 24.20 28.77 33.01 32.87 31.93 33.54 0.8323 0.9129 0.9537 0.9056 0.9735 0.9526 Intersection 23.77 31.23 35.51 34.35 35.52 35.82 0.8548 0.9426 0.9770 0.9698 0.9463 0.9788 Mediumresidential 25.54 26.78 26.89 26.85 27.03 29.06 0.8059 0.8292 0.8385 0.7150 0.8350 0.8810 Mobilehomepark 21.40 24.00 24.59 25.66 25.31 26.52 0.7436 0.8422 0.8705 0.8317 0.8885 0.8905 Overpass 25.74 29.59 29.10 31.35 32.70 34.80 0.8366 0.9202 0.9156 0.8876 0.9543 0.9671 Parkinglot 25.04 25.83 29.65 29.30 30.23 33.98 0.8644 0.8830 0.9363 0.8807 0.9455 0.9700 River 25.90 28.04 28.18 27.92 28.53 29.83 0.7772 0.8272 0.8346 0.6820 0.8422 0.6940 Runway 23.74 36.45 38.82 37.04 36.08 39.22 0.8125 0.9633 0.9737 0.9173 0.9701 0.9751 Sparseresidential 25.61 26.75 26.58 26.36 26.53 26.82 0.7199 0.8020 0.8136 0.6676 0.8033 0.6680 Storagetanks 25.66 23.83 23.99 25.62 25.85 27.23 0.7973 0.8106 0.8344 0.7700 0.8723 0.9235 Tenniscourt 27.83 29.43 31.85 31.25 30.53 32.63 0.8861 0.9104 0.9280 0.8807 0.9402 0.9386
Table 2. Ablation experimental results
View tableTable 2. Ablation experimental results
PSNR /dB SSIM — 29.03 0.8345 √ 32.06 0.8969
Tools
Get Citation
Copy Citation Text
Lijing Bu, Beini Yang, Guoqiang Dong, Zhengpeng Zhang, Yin Yang, Yujie Feng. Interpretable Deep Learning Image Restoration Algorithm with L2-Norm Prior[J]. Laser & Optoelectronics Progress, 2025, 62(6): 0637004
Paper Information
Category: Digital Image Processing
Received: May. 23, 2024
Accepted: Jul. 29, 2024
Published Online: Mar. 5, 2025
The Author Email:
CSTR:32186.14.LOP241353
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20