Hybrid distortion image correction method based on improved U-Net networks

	$f_{x}$	$f_{y}$	$c_{x}$	$c_{y}$	$k_{1}$	$k_{2}$	$k_{3}$	$p_{1}$	$p_{2}$
1	219.806	293.039	255.107	253.441	-0.219 67	0.057 86	-0.008 06	-0.000 46	-0.000 37
2	223.262	297.586	258.437	255.308	-0.261 51	0.095 77	-0.018 77	-0.000 37	0.000 17
3	302.224	399.584	256.349	254.486	-0.279 23	0.133 74	-0.044 35	0.000 13	0.000 13
4	451.199	596.817	259.112	255.498	-0.282 33	0.159 47	-0.083 70	-0.000 99	0.000 02

Table 2. Evaluation results of different methods on the test dataset

View table

View in Article

Table 2. Evaluation results of different methods on the test dataset

Methods	MAE	PSNR	SSIM
Li^［14］	1.442 0	34.751 7	0.832 5
Hosono^［25］	N/A	33.494 2	0.633 2
U-Net^［21］	0.336 1	35.747 1	0.924 3
PSPNet^［26］	1.18	34.786 7	0.845 9
DeepLabV3+^［27］	0.978 5	34.982 2	0.867 6
Ours	0.251 9	35.796 8	0.930 6

Table 3. Description of the grading scale

View table

View in Article

Table 3. Description of the grading scale

评分等级	评分说明
5	与参考图像相比，感受不到差异。
4	与参考图像相比，感受到轻微差异。
3	与参考图像相比，感受到较大差异。
2	与参考图像相比，感受到很大差异。
1	与参考图像相比，感受到非常大差异。

Table 4. Subjective evaluation scores of different methods
View table
View in Article
Table 4. Subjective evaluation scores of different methods
Li Hosono U-Net PSPNet DeepLabV3+ Ours
MOS 3.722 1.967 4.339 3.117 3.017 4.597
Variance 0.962 0.943 0.674 1.059 1.128 0.385

Table 5. Ablation experiments of loss function
View table
View in Article
Table 5. Ablation experiments of loss function
EPE $L_{0}$ $L_{1}$ $L_{2}$ MAE PSNR SSIM
√ × × × 1.123 9 35.046 6 0.854 3
× √ × × 0.377 0 35.628 1 0.918 3
× × √ × 0.281 0 35.772 0 0.926 2
× × √ √ 0.251 9 35.796 8 0.930 6

Tools

Get Citation

Copy Citation Text

Wei SONG, Li-biao SHI, Li-jia GENG, Zhen-ling MA, Yan-ling DU. Hybrid distortion image correction method based on improved U-Net networks[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(11): 1580

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites