Chinese Journal of Liquid Crystals and Displays, Volume. 37, Issue 4, 508(2022)
Technical research of composite residual network in low illumination image enhancement
Fig. 5. Influence of different loss weights on experimental results
Fig. 6. Visual subjective contrast on synthetic low illumination images.（a）Ship；（b）Indoor scenes；（c）Airplane；（d）Bus；（e）Roller coaste.
Fig. 7. Visual subjective comparison of real low illumination images in LOL dataset.（a）Sculpture 1；（b）Clock；（c）House；（d）Sculpture 2；（e）Trees.
Fig. 8. Visual subjective comparison of real low illumination images in Exdark dataset.（a）Dragon boat；（b）Castle；（c）People；（d）Road；（e）Room.
Table 1. Comparison of evaluation indicators in Fig.5
View tableView in ArticleTable 1. Comparison of evaluation indicators in Fig.5
Evaluation
indicator
0.9-0.1 0.1-0.9 0.7-0.3 House PSNR/dB 15.821 14.586 19.354 SSIM 0.581 0.42 0.853 MSE 1 702.241 2 262.009 977.44 Tree PSNR/dB 19.9585 16.957 22.459 SSIM 0.8514 0.686 0.846 MSE 656.4912 1 330.356 684.981 Structure PSNR/dB 21.5879 19.282 22.524 SSIM 0.8648 0.774 0.865 MSE 451.1169 767.206 449.708
Table 2. Comparison of objective evaluation indexes of each image in Fig.6
View tableView in ArticleTable 2. Comparison of objective evaluation indexes of each image in Fig.6
Image Evaluation indicator HE aMSRCR MSRCP LIME LNet Our Ship image PSNR/dB 14.532 16.744 12.119 11.750 20.968 21.497 SSIM 0.848 0.894 0.778 0.711 0.955 0.956 MSE 2 290.231 1 376.139 3 991.753 4 345.288 520.263 498.322 Indoor scenes image PSNR/dB 19.158 15.820 11.256 13.921 20.103 21.117 SSIM 0.835 0.756 0.744 0.743 0.920 0.917 MSE 789.322 1 702.462 4 868.392 2 635.626 634.940 611.056 Airplane image PSNR/dB 16.385 13.500 12.203 12.768 13.374 19.017 SSIM 0.604 0.413 0.839 0.751 0.894 0.955 MSE 1 494.625 2 904.467 3 915.226 3 437.761 2 989.908 1 892.217 Bus image PSNR/dB 23.960 15.451 7.848 15.872 22.036 21.746 SSIM 0.879 0.822 0.535 0.825 0.903 0.897 MSE 261.250 1 853.029 10 670.485 1 682.118 406.809 492.947 Roller coaster image PSNR/dB 12.039 10.582 12.039 12.833 20.423 22.060 SSIM 0.641 0.599 0.641 0.737 0.910 0.932 MSE 4 065.351 4 282.467 4 065.351 4 263.441 589.814 451.394
Table 3. Comparison of objective evaluation indexes of each image in Fig.7
View tableView in ArticleTable 3. Comparison of objective evaluation indexes of each image in Fig.7
Image Evaluation indicator HE aMSRCR MSRCP LIME LNet Our Sculpture1 PSNR/dB 16.072 18.601 17.164 19.295 24.608 25.044 SSIM 0.709 0.798 0.872 0.895 0.923 0.930 MSE 2 280.106 1 790.273 1 859.271 1 344.615 225.013 207.573 Clock PSNR/dB 12.097 12.516 15.461 16.112 20.082 25.022 SSIM 0.726 0.615 0.852 0.863 0.876 0.914 MSE 4 012.073 3 642.859 1 848.916 1 591.709 637.949 204.582 House PSNR/dB 18.417 16.594 17.545 13.353 14.448 19.354 SSIM 0.846 0.795 0.836 0.731 0.665 0.853 MSE 936.074 1 424.478 1 144.366 3 004.395 2 334.539 977.44 Sculpture2 PSNR/dB 21.392 15.116 10.625 25.297 13.639 22.524 SSIM 0.849 0.707 0.748 0.874 0.793 0.865 MSE 471.932 2 001.877 5 630.381 692.029 2 812.742 449.708 Trees PSNR/dB 13.791 14.343 13.732 16.233 18.788 22.459 SSIM 0.758 0.654 0.705 0.835 0.827 0.876 MSE 2 715.952 2 391.888 2 752.863 1 547.803 859.374 684.981
Table 4. Comparison of objective evaluation indexes of each image in Fig.8
View tableView in ArticleTable 4. Comparison of objective evaluation indexes of each image in Fig.8
Image Evaluation indicator HE aMSRCR MSRCP LIME LNet Our Dragon boat Average 58.102 123.259 87.486 41.762 28.204 57.208 Standard Deviation 35.407 44.261 41.455 30.579 29.896 46.552 Castle Average 68.753 122.313 147.918 85.608 83.602 136.038 Standard Deviation 42.357 70.236 38.462 35.255 44.971 57.627 People Average 67.002 120.490 108.001 49.602 33.994 111.076 Standard Deviation 36.959 39.316 41.078 30.075 32.013 42.216 Road Average 78.965 129.138 134.429 65.990 59.219 83.170 Standard Deviation 49.023 58.118 52.490 37.420 39.636 51.705 Room Average 90.558 124.962 141.688 78.835 86.300 118.723 Standard Deviation 48.626 59.647 52.334 41.589 49.473 56.793
Table 5. Comparison of average running time of each algorithm
View tableView in ArticleTable 5. Comparison of average running time of each algorithm
Algorithm Time/s HE 0.349 aMSRCR 0.483 MSRCP 0.900 LIME 27.539 LNet 5.791 Our 2.575
Get Citation
Copy Citation Text
Xing-rui WANG, Yan PIAO, Yu-mo WANG. Technical research of composite residual network in low illumination image enhancement[J]. Chinese Journal of Liquid Crystals and Displays, 2022, 37(4): 508
Category:
Received: Aug. 30, 2021
Accepted: --
Published Online: Jun. 20, 2022
The Author Email: Yan PIAO (piaoyan@cust.edu.cn)
