基于多先验约束的雾霾图像复原 下载: 866次
Haze Image Restoration Based on Multi-Prior Constraints
1 西安财经大学管理学院, 陕西 西安 710100
2 空军工程大学航空工程学院, 陕西 西安 710038
图 & 表
图 1. 大气散射模型
Fig. 1. Atmospheric scattering model
下载图片 查看原文
图 2. “远山”与“红楼”颜色衰减先验的分析结果。(a)雾霾图像;(b)黑白差值图;(c)亮度与饱和度的差值与场景深度的关系曲线
Fig. 2. Analysis results of color attenuation prior of “distant mountain” and “red building”. (a) Haze images; (b) black and white difference images; (c) relationship curves between brightness and saturation difference and scene depth
下载图片 查看原文
图 3. 不同算法处理“远山”、“道路”和“楼群”雾霾图像及相应的可见边缘。(a)(b)输入的雾霾图像;(c)(d)文献[
5];(e)(f)所提算法
Fig. 3. Different algorithms process “distant mountain”, “road” and “building” smog images and corresponding visible edges. (a)(b) Input haze images; (c)(d) Ref. [5]; (e)(f) proposed algorithm
下载图片 查看原文
图 4. 不同算法处理“麦草”、“山丘”及“橙子”雾霾图像及相应的可见边缘。(a)(b)输入雾霾图像;(c)(d)文献[
12];(e)(f)所提算法
Fig. 4. Different algorithms process “wheatgrass”, “hill” and “orange” haze images and corresponding visible edges. (a)(b) Input haze images; (c)(d) Ref. [12]; (e)(f) proposed algorithm
下载图片 查看原文
图 5. 不同算法处理“街道”、“林间”和“村庄”雾霾图像及相应的可见边缘。 (a)(b)输入雾霾图像;(c)(d)文献[
13];(e)(f)所提算法
Fig. 5. Different algorithms process “street”, “forest” and “village” smog images and corresponding visible edges. (a)(b) Input haze images; (c)(d) Ref. [13]; (e)(f) proposed algorithm
下载图片 查看原文
图 6. 不同算法处理“鹅”、“小楼”和“山林”雾霾图像及相应的可见边缘。(a)(b)输入雾霾图像;(c)(d)文献[
8];(e)(f)所提算法
Fig. 6. Different algorithms process “goose”, “small building” and “mountain forest” smog images and corresponding visible edges. (a)(b) Input haze images; (c)(d) Ref. [8]; (e)(f) proposed algorithm
下载图片 查看原文
表 1不同算法处理图3的VDI和HR对比
Table1. Comparison of VDI and HR in Fig. 3 with different algorithms
Fig.3 | VDI | HR |
---|
Ref. [5] | Proposedalgorithm | Ref. [5] | Proposedalgorithm |
---|
Distantmountain | 0.4376 | 0.5832 | 0.6817 | 0.8673 | Road | 0.3391 | 0.6997 | 0.4538 | 0.7261 | Building | 0.5248 | 0.6613 | 0.6254 | 0.8392 |
|
查看原文
表 2不同算法处理图4的VDI和HR对比
Table2. Comparison of VDI and HR in Fig. 4 with different algorithms
Fig.4 | VDI | HR |
---|
Ref. [12] | Proposedalgorithm | Ref. [12] | Proposedalgorithm |
---|
Wheatgrass | 0.3194 | 0.6582 | 0.5833 | 0.7761 | Hill | 0.5288 | 0.7917 | 0.6702 | 0.8714 | Orange | 0.4570 | 0.6746 | 0.6091 | 0.8418 |
|
查看原文
表 3不同算法处理图5的VDI和HR对比
Table3. Comparison of VDI and HR in Fig. 5 with different algorithms
Fig.5 | VDI | HR |
---|
Ref. [13] | Proposedalgorithm | Ref. [13] | Proposedalgorithm |
---|
Street | 0.4277 | 0.7301 | 0.6375 | 0.8399 | Forest | 0.4591 | 0.6228 | 0.6027 | 0.7924 | Village | 0.4370 | 0.6419 | 0.6518 | 0.9067 |
|
查看原文
表 4不同算法处理图6的VDI和HR对比
Table4. Comparison of VDI and HR in Fig. 6 with different algorithms
Fig.6 | VDI | HR |
---|
Ref. [8] | Proposedalgorithm | Ref. [8] | Proposedalgorithm |
---|
Goose | 0.4136 | 0.6813 | 0.5195 | 0.7367 | Smallbuilding | 0.5288 | 0.7291 | 0.6830 | 0.8166 | Mountainforest | 0.4930 | 0.6205 | 0.4729 | 0.7804 |
|
查看原文
曲晨, 毕笃彦. 基于多先验约束的雾霾图像复原[J]. 激光与光电子学进展, 2020, 57(18): 181014. Chen Qu, Duyan Bi. Haze Image Restoration Based on Multi-Prior Constraints[J]. Laser & Optoelectronics Progress, 2020, 57(18): 181014.