基于多尺度Retinex和暗通道的自适应图像去雾算法 下载: 890次
Adaptive Image Defogging Algorithm Combining Multi-Scale Retinex and Dark Channel
兰州交通大学电子与信息工程学院, 甘肃 兰州 730070
图 & 表
图 1. 所提算法的处理流程
Fig. 1. Flowchart of proposed algorithm
下载图片 查看原文
图 2. 不同方法处理后的透射率图像。(a)有雾图像;(b)文献[
8];(c)透射率粗估计;(d)优化透射率
Fig. 2. Transmittance images after processing by different methods. (a) Hazy images; (b) Ref. [8]; (c) rough estimations of transmittance; (d) transmittance optimization
下载图片 查看原文
图 3. 不同算法处理近景图像的效果。(a)原图;(b)文献[
7];(c) 文献[
8]; (d) 文献[
9]; (e) 文献[
12];(f)所提算法
Fig. 3. Effect of different algorithms for processing close-range images. (a) Original images; (b) Ref. [7]; (c) Ref. [8]; (d) Ref. [9]; (e) Ref. [12]; (f) proposed algorithm
下载图片 查看原文
图 4. 不同算法处理远近交替图像的效果。(a)原图;(b)文献[
7];(c) 文献[
8]; (d) 文献[
9]; (e) 文献[
12];(f)所提算法
Fig. 4. Effect of different algorithms for processing alternating near-and-far images. (a) Original images; (b) Ref. [7]; (c) Ref. [8]; (d) Ref. [9]; (e) Ref. [12]; (f) proposed algorithm
下载图片 查看原文
图 5. 不同算法处理远景图像的效果。(a)原图;(b)文献[
7];(c) 文献[
8]; (d) 文献[
9]; (e) 文献[
12];(f)所提算法
Fig. 5. Effect of different algorithms for processing perspective images. (a) Original images; (b) Ref. [7]; (c) Ref. [8]; (d) Ref. [9]; (e) Ref. [12]; (f) proposed algorithm
下载图片 查看原文
表 1不同算法处理不同图像的p值
Table1. p values of processing different images by different algorithms
Image | Ref. [7] | Ref. [8] | Ref. [9] | Ref. [12] | Proposed algorithm |
---|
1 | 0.0997 | 0.0867 | 0.0318 | 0.0148 | 0.1143 | 2 | 0.2492 | 0.2994 | 0.1021 | 0.1775 | 0.4928 | 3 | 0.3485 | 0.3967 | -0.0021 | 0.1280 | 0.3821 | 4 | 0.3115 | 0.1456 | -0.0789 | 0.0553 | 0.2071 | 5 | 0.0852 | -0.0266 | -0.1392 | 0.0849 | 0.1729 | 6 | 0.1451 | 0.0823 | -0.0070 | 0.0627 | 0.1423 | 7 | -0.0350 | -0.0150 | -0.2082 | 0.0412 | 0.1095 | Mean | 0.1720 | 0.1384 | -0.0431 | 0.0806 | 0.2316 |
|
查看原文
表 2不同算法处理不同图像的t值
Table2. t values of processing different images by different algorithms
Image | Ref. [7] | Ref. [8] | Ref. [9] | Ref. [12] | Proposed algorithm |
---|
1 | 1.2296 | 1.1788 | 1.0159 | 1.0657 | 1.3386 | 2 | 1.1138 | 1.1596 | 1.0685 | 1.2269 | 1.6675 | 3 | 1.1859 | 1.1761 | 1.1169 | 1.1751 | 1.5083 | 4 | 1.1104 | 1.0774 | 0.9077 | 1.1679 | 1.6777 | 5 | 1.1814 | 1.0341 | 0.8125 | 1.1625 | 1.5929 | 6 | 1.1371 | 1.0750 | 1.0229 | 1.1397 | 1.3475 | 7 | 1.0939 | 1.1230 | 0.9799 | 1.1159 | 1.8649 | Mean | 1.1503 | 1.1177 | 0.9892 | 1.1505 | 1.5711 |
|
查看原文
表 3不同算法处理不同图像的PSNR
Table3. PSNR of processing different images by different algorithms
Image | Ref. [7] | Ref. [8] | Ref. [9] | Ref. [12] | Proposed algorithm |
---|
1 | 61.2513 | 61.4723 | 62.1003 | 65.6993 | 62.1086 | 2 | 55.8298 | 57.0069 | 58.6501 | 58.6995 | 63.4518 | 3 | 58.0261 | 58.0033 | 61.2204 | 64.6648 | 61.8904 | 4 | 59.1774 | 55.5100 | 60.1118 | 65.4617 | 63.4308 | 5 | 60.9005 | 59.6184 | 61.0325 | 67.0162 | 62.3975 | 6 | 60.5424 | 59.5899 | 61.4846 | 68.1238 | 63.3756 | 7 | 59.0992 | 60.1070 | 61.4118 | 66.0904 | 60.2116 | Mean | 59.2610 | 58.7583 | 60.8588 | 65.1080 | 62.4095 |
|
查看原文
表 4不同算法处理不同图像的运行时间
Table4. Running time of processing different images by different algorithms unit: s
Image | Ref. [7] | Ref. [8] | Ref. [9] | Ref. [12] | Proposed algorithm |
---|
1 | 0.5069 | 0.0932 | 2.5809 | 1.5018 | 1.1069 | 2 | 1.1512 | 0.0841 | 2.2040 | 1.7727 | 1.1731 | 3 | 0.5956 | 0.2506 | 1.4160 | 3.8809 | 2.3102 | 4 | 0.4769 | 0.1391 | 1.5732 | 1.9901 | 1.2699 | 5 | 0.4351 | 0.0484 | 1.6150 | 0.7655 | 0.7919 | 6 | 0.4399 | 0.0864 | 1.9004 | 1.3791 | 1.0923 | 7 | 0.4036 | 0.0580 | 1.0559 | 0.6043 | 0.5253 | Mean | 0.5727 | 0.1085 | 1.7636 | 1.6992 | 1.1814 |
|
查看原文
彭静, 薛奉金, 苑玉彬. 基于多尺度Retinex和暗通道的自适应图像去雾算法[J]. 激光与光电子学进展, 2021, 58(4): 0410004. Jing Peng, Fengjin Xue, Yubin Yuan. Adaptive Image Defogging Algorithm Combining Multi-Scale Retinex and Dark Channel[J]. Laser & Optoelectronics Progress, 2021, 58(4): 0410004.