A coal mine underground image enhancement method based on multi-scale gradient domain guided image filtering
-
摘要: 煤矿井下图像存在较严重的光照不均匀和噪声干扰，现有基于Retinex的方法直接应用于煤矿井下图像增强易出现光晕伪影、边缘模糊、过增强和噪声放大等问题。针对上述问题，提出了一种基于多尺度梯度域引导滤波的煤矿井下图像增强方法。首先，将多尺度思想引入梯度域引导滤波中，实现对非均匀光照的准确估计，有效解决了增强图像时光晕伪影及边缘模糊的问题。然后，利用Retinex模型分离出光照分量和反射分量：对于光照分量，通过自适应伽马校正函数逐像素地修正光照信息，实现对图像暗区域增强的同时，抑制亮区域过增强，并使用限制对比度自适应直方图均衡化方法调整图像对比度；对于反射分量，将梯度域引导滤波与多尺度细节提升相结合，在准确去除噪声后提升纹理细节，避免了增强图像时噪声放大的问题。最后，将处理后的光照分量及反射分量融合，计算图像增益系数，并使用线性色彩恢复方法实现对原始RGB图像的逐像素增强，提升方法处理效率。实验结果表明，从主客观角度与现有方法相比，经所提方法处理后的图像在色彩保持、对比度、噪声抑制、细节保留等方面均取得了较好的增强效果，同时处理效率较高。Abstract: There are serious issues with uneven lighting and noise interference in coal mine underground images. The existing Retinex based methods are directly applied to enhance coal mine underground images, which are prone to problems such as halo artifacts, blurred edges, over enhancement, and noise amplification. In order to solve the above problems, a coal mine underground image enhancement method based on multi-scale gradient domain guided image filtering is proposed. Firstly, the multi-scale idea is introduced into gradient domain guided image filtering to achieve accurate estimation of non-uniform lighting, effectively solving the problems of halo artifacts and edge blurring in enhanced images. Secondly, the Retinex model is used to separate the lighting component and reflection component. For the lighting component, the lighting information is corrected pixel by pixel through an adaptive gamma correction function, which enhances the dark areas of the image while suppressing the over enhancement of the bright areas. The image contrast is adjusted using a contrast limited adaptive histogram equalization method. For the reflection component, gradient domain guided image filtering is combined with multi-scale detail enhancement to accurately remove noise and improve texture details, avoiding the problem of noise amplification during image enhancement. Finally, the processed lighting and reflection components are fused, and the image gain coefficient is calculated. The linear color restoration method is used to enhance the original RGB image pixel by pixel, improving the processing efficiency of the method. The experimental results show that, from a subjective and objective perspective, compared with existing methods, the images processed by the proposed method have achieved better enhancement effects in color preservation, contrast, noise suppression, detail preservation, and other aspects, while also having higher processing efficiency.
-
-
![]()
图 1 基于多尺度GDGIF的煤矿井下图像增强方法流程
Figure 1. Flow of coal mine underground images enhancement method based on multi-scale gradient domain guided filtering (GDGIF)
![]()
图 7 光照分量处理结果主观对比
Figure 7. Subjective comparison of lighting component processing results
![]()
图 8 反射分量处理结果主观对比
Figure 8. Subjective comparison of reflection component processing results
![]()
图 9 采用线性色彩恢复前后运行时间对比
Figure 9. Comparison of running time before and after linear color restoration
表 1 光照不足图像增强后的客观指标对比
Table 1 Comparison of objective indicators of insufficient lighting images after enhancement
图像 方法 熵 对比度变化率 色调变化率 能量梯度/1011 方差/109 自相关函数/109 场景1 原图 3.7006 − − 0.0285 0.0651 0.0550 MSR方法 5.8865 24.8618 0.0281 4.0457 1.6612 1.5083 MSRCR方法 6.5680 8.7692 1.1135 3.1408 0.6210 0.4817 NPE方法 6.0668 32.4399 0.0578 7.3827 2.3033 2.0880 文献[13]方法 5.3232 11.7606 0.0025 3.7495 0.8331 0.6932 SRLIE方法 6.3004 7.1466 0.0819 0.6554 0.5423 0.4789 文献[16]方法 5.4670 9.4856 0.0071 3.3099 0.6775 0.5279 本文方法 6.0595 27.1673 0.0170 18.4625 1.8830 1.6420 场景2 原图 2.2705 − − 0.0142 0.0419 0.0395 MSR方法 4.6642 36.6351 0.1440 3.4645 1.5758 1.4915 MSRCR方法 6.8957 29.4952 1.0857 9.3618 1.2672 1.0972 NPE方法 4.5409 72.8817 0.0742 12.4535 3.1708 3.0054 文献[13]方法 3.7798 20.1045 0.0607 4.3327 0.8909 0.8446 SRLIE方法 5.5819 8.8160 0.0736 0.4410 0.4157 0.4010 文献[16]方法 3.9513 9.7210 0.0700 2.2207 0.4481 0.4014 本文方法 4.4048 38.0720 0.0988 14.9444 1.6573 1.5196 场景3 原图 6.1595 − − 0.6860 0.2941 0.2667 MSR方法 7.3049 0.6172 0.0230 3.7913 0.5044 0.4251 MSRCR方法 7.4457 1.4760 0.4080 6.6102 0.7489 0.6135 NPE方法 7.4347 0.8684 0.0052 2.9075 0.5706 0.4758 文献[13]方法 6.9556 1.0378 0.0267 3.3654 0.6071 0.5372 SRLIE方法 7.2143 0.7327 0.0013 6.0787 0.5241 0.4243 文献[16]方法 7.2542 0.9368 0.0069 5.8487 0.5859 0.4777 本文方法 7.5721 2.1572 0.0060 15.1204 0.9552 0.7847 
下载: 导出CSV
表 2 光照不均匀图像增强后的客观指标对比
Table 2 Comparison of objective indicators of uneven lighting images after enhancement
图像 方法 熵 对比度变化率 色调变化率 能量梯度/1011 方差/109 自相关函数/109 场景4 原图 6.3582 − − 0.1057 0.7753 0.7626 MSR方法 6.8935 −0.3635 0.1976 0.0276 0.4907 0.4531 MSRCR方法 7.2880 0.7271 0.4387 0.9089 1.3167 1.2255 NPE方法 7.1122 −0.0713 0.1279 0.2362 0.7256 0.6748 文献[13]方法 6.8919 0.4901 0.0815 0.2372 1.1574 1.1317 SRLIE方法 6.9172 0.0049 0.1109 0.4741 0.7981 0.7585 文献[16]方法 7.0076 0.1236 0.1041 0.4953 0.8753 0.8257 本文方法 7.3756 0.7282 0.1104 0.1477 1.3624 1.2867 场景5 原图 6.8731 − − 4.1316 2.0520 1.9688 MSR方法 7.4342 −0.2985 0.0049 6.9938 2.0619 1.9503 MSRCR方法 7.5521 −0.3144 0.3124 13.6200 1.4388 1.2524 NPE方法 7.3378 −0.3520 0.0035 3.6159 1.5882 1.4974 文献[13]方法 7.1349 −0.0272 0.0032 5.4609 2.0646 1.9621 SRLIE方法 7.4616 −0.2029 0.0041 14.7030 2.0804 1.8946 文献[16]方法 7.3964 0.0362 0.0057 11.6369 2.3043 2.2220 本文方法 7.6240 −0.0095 0.0053 21.2435 2.3077 2.0989 场景6 原图 6.9058 − − 2.1236 1.2952 1.1667 MSR方法 7.3849 −0.2688 0.0278 4.2159 0.9806 0.8356 MSRCR方法 7.5106 0.1842 0.0590 5.6076 1.4880 1.4880 NPE方法 7.5147 −0.0426 0.0021 2.9998 1.2639 1.1009 文献[13]方法 7.2822 0.4765 0.0146 3.8806 1.9159 1.7521 SRLIE方法 7.5766 0.1840 0.0182 15.0448 1.5822 1.2868 文献[16]方法 7.6350 0.3503 0.0071 9.6458 1.7720 1.4997 本文方法 7.6898 0.6975 0.0046 15.1745 2.2122 1.9108
下载: 导出CSV
表 3 有噪声图像增强后的客观指标对比
Table 3 Comparison of objective indicators of noisy images after enhancement
图像 方法 熵 对比度变化率 色调变化率 能量梯度/1011 方差/109 自相关函数/109 场景7 原图 7.4796 − − 1.1089 1.9205 1.8342 MSR方法 6.7098 −0.6366 0.1581 0.9117 0.6847 0.5985 MSRCR方法 7.5967 0.1879 0.3258 12.7740 2.3104 2.0120 NPE方法 7.3893 −0.2476 0.0021 1.2927 1.4258 1.3161 文献[13]方法 7.5420 0.0686 0.0032 1.3811 2.0567 1.9576 SRLIE方法 7.4700 −0.1171 0.0010 6.5847 1.6743 1.4649 文献[16]方法 7.5153 −0.0468 0.0022 4.8278 1.8161 1.6207 本文方法 7.6492 0.0988 0.0008 8.1902 2.1155 1.8958 场景8 原图 7.5702 − − 0.8719 0.5965 0.6024 MSR方法 7.2635 −0.4969 0.0956 2.1923 0.3936 0.3317 MSRCR方法 7.6873 −0.2781 0.7638 7.3505 0.5826 0.4513 NPE方法 7.5994 −0.2304 0.0268 1.2377 0.6167 0.5484 文献[13]方法 7.6109 −0.0994 0.0616 1.1409 0.7187 0.6616 SRLIE方法 7.7672 −0.0684 0.0135 6.7768 0.7477 0.6147 文献[16]方法 7.7835 −0.0220 0.0134 4.4739 0.7822 0.6681 本文方法 7.8190 0.0970 0.0062 9.6410 0.8761 0.7370
下载: 导出CSV
表 4 不同方法处理图像的平均时间对比
Table 4 Comparison of average time of images processed by different methods
s
下载: 导出CSV
表 5 光照分量处理结果客观对比
Table 5 Objective comparison of lighting component processing results
方法 熵 对比度变化率 色调变化率 能量梯度/1011 方差/109 自相关函数/109 原图 7.3510 − − 1.4961 1.5800 1.4998 仅使用自适应伽马校正 7.7851 0.3955 0.0069 6.8291 2.2033 2.0339 仅使用CLAHE 7.5874 1.4987 0.0076 13.4309 3.2212 3.6376 自适应伽马校正+CLAHE 7.8128 1.3546 0.0084 14.7554 3.3903 3.1379
下载: 导出CSV
表 6 反射分量处理结果客观对比
Table 6 Objective comparison of reflection component processing results
方法 熵 对比度变化率 色调变化率 能量梯度/1011 方差/109 自相关函数/109 原图 7.4236 − − 1.0021 0.6663 0.6130 仅使用GDGIF去噪 7.6786 0.3246 0.0166 3.6185 0.8806 0.7758 仅使用多尺度细节提升 7.7891 0.6205 0.0181 12.0516 1.0743 0.8447 GDGIF去噪+多尺度细节提升 7.7903 0.6280 0.0154 10.8591 1.0949 0.8569
下载: 导出CSV
-
[1] 王国法,刘峰,庞义辉,等. 煤矿智能化——煤炭工业高质量发展的核心技术支撑[J]. 煤炭学报,2019,44(2):349-357. WANG Guofa,LIU Feng,PANG Yihui,et al. Coal mine intellectualization:the core technology of high quality development[J]. Journal of China Coal Society,2019,44(2):349-357.
[2] 张立亚,郝博南,孟庆勇,等. 基于HSV空间改进融合Retinex算法的井下图像增强方法[J]. 煤炭学报,2020,45(增刊1):532-540. ZHANG Liya,HAO Bonan,MENG Qingyong,et al. Method of image enhancement in coal mine based on improved retex fusion algorithm in HSV space[J]. Journal of China Coal Society,2020,45(S1):532-540.
[3] 程德强,钱建生,郭星歌,等. 煤矿安全生产视频AI识别关键技术研究综述[J]. 煤炭科学技术,2023,51(2):349-365. CHENG Deqiang,QIAN Jiansheng,GUO Xingge,et al. Review on key technologies of AI recognition for videos in coal mine[J]. Coal Science and Technology,2023,51(2):349-365.
[4] 张谢华,张申,方帅,等. 煤矿智能视频监控中雾尘图像的清晰化研究[J]. 煤炭学报,2014,39(1):198-204. ZHANG Xiehua,ZHANG Shen,FANG Shuai,et al. Clearing research on fog and dust images in coalmine intelligent video surveillance[J]. Journal of China Coal Society,2014,39(1):198-204.
[5] 王国法,张良,李首滨,等. 煤矿无人化智能开采系统理论与技术研发进展[J]. 煤炭学报,2023,48(1):34-53. WANG Guofa,ZHANG Liang,LI Shoubin,et al. Progresses in theory and technological development of unmanned smart mining system[J]. Journal of China Coal Society,2023,48(1):34-53.
[6] LI Chongyi,GUO Chunle,HAN Linghao,et al. Lighting the darkness in the deep learning era[Z]. 2021. DOI: 10.48550/arXiv.2104.10729.
[7] 王焱,关南楠,刘海涛. 改进的多尺度Retinex井下图像增强算法[J]. 辽宁工程技术大学学报(自然科学版),2016,35(4):440-443. DOI: 10.11956/j.issn.1008-0562.2016.04.020 WANG Yan,GUAN Nannan,LIU Haitao. An improved multi-scale Retinex algorithm for mine image enhancement[J]. Journal of Liaoning Technical University(Natural Science),2016,35(4):440-443. DOI: 10.11956/j.issn.1008-0562.2016.04.020
[8] 甘建旺. 矿井图像的去噪与增强算法研究[D]. 北京:北京石油化工学院,2021. GAN Jianwang. Research on denoising and enhancement algorithm of mine image[D]. Beijing:Beijing Institute of Petrochemical Technology,2021.
[9] JOBSON D J,RAHMAN Z,WOODELL G A. Properties and performance of a center/surround Retinex[J]. IEEE Transactions on Image Processing,1997,6(3):451-462. DOI: 10.1109/83.557356
[10] JOBSON D J,RAHMAN Z. A multiscale Retinex for bridging the gap between color images and the human observation of scenes[J]. IEEE Transactions on Image Processing,1997,6(7):965-976. DOI: 10.1109/83.597272
[11] RAHMAN Z,JOBSON D,WOODELL G A. Retinex processing for automatic image enhancement[J]. Journal of Electronic Imaging,2004,13(1):100-110. DOI: 10.1117/1.1636183
[12] 程德强,郑珍,姜海龙. 一种煤矿井下图像增强算法[J]. 工矿自动化,2015,41(12):31-34. CHENG Deqiang,ZHENG Zhen,JIANG Hailong. An image enhancement algorithm for coal mine underground[J]. Industry and Mine Automation,2015,41(12):31-34.
[13] 智宁,毛善君,李梅. 基于双伽马函数的煤矿井下低亮度图像增强算法[J]. 辽宁工程技术大学学报(自然科学版),2018,37(1):191-197. DOI: 10.11956/j.issn.1008-0562.2018.01.034 ZHI Ning,MAO Shanjun,LI Mei. An enhancement algorithm for coal mine low illumination images based on bi-gamma function[J]. Journal of Liaoning Technical University(Natural Science),2018,37(1):191-197. DOI: 10.11956/j.issn.1008-0562.2018.01.034
[14] HE Kaiming,SUN Jian,TANG Xiaoou. Guided image filtering[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2012,35(6):1397-1409.
[15] 王星,白尚旺,潘理虎,等. 一种矿井图像增强算法[J]. 工矿自动化,2017,43(3):48-52. WANG Xing,BAI Shangwang,PAN Lihu,et al. A mine image enhancement algorithm[J]. Industry and Mine Automation,2017,43(3):48-52.
[16] 洪炎,朱丹萍,龚平顺. 基于TopHat加权引导滤波的Retinex矿井图像增强算法[J]. 工矿自动化,2022,48(8):43-49. HONG Yan,ZHU Danping,GONG Pingshun. Retinex mine image enhancement algorithm based on TopHat weighted guided filtering[J]. Journal of Mine Automation,2022,48(8):43-49.
[17] 苏波,李超,王莉. 基于多权重融合策略的Retinex矿井图像增强算法[J]. 煤炭学报,2023,48(增刊2):813-822. SU Bo,LI Chao,WANG Li. Mine image enhancement algorithm based on Retinex using multi-weight fusion strategy[J]. Journal of China Coal Society,2023,48(S2):813-822.
[18] KOU Fei,CHEN Weihai,WEN Changyun,et al. Gradient domain guided image filtering[J]. IEEE Transactions on Image Processing,2015,24(11):4528-4539. DOI: 10.1109/TIP.2015.2468183
[19] 马龙,马腾宇,刘日升. 低光照图像增强算法综述[J]. 中国图象图形学报,2022,27(5):1392-1409. DOI: 10.11834/jig.210852 MA Long,MA Tengyu,LIU Risheng. The review of low-light image enhancement[J]. Journal of Image and Graphics,2022,27(5):1392-1409. DOI: 10.11834/jig.210852
[20] WEI Chen,WANG Wenjing,YANG Wenhan,et al. Deep Retinex decomposition for low-light enhancement[Z]. 2018. DOI: 10.48550/arXiv.1808.04560.
[21] ZHANG Yonghua,ZHANG Jiawan,GUO Xiaojie. Kindling the darkness:a practical low-light image enhancer[C]. The 27th ACM International Conference on Multimedia,Nice,2019:1632-1640.
[22] JIANG He,CAI Huangkai,YANG Jie. Learning in-place residual homogeneity for image detail enhancement[C]. IEEE International Conference on Acoustics,Speech and Signal Processing,Calgary,2018:1428-1432.
[23] JIANG He,ASAD M,LIU Jingjing,et al. Single image detail enhancement via metropolis theorem[J]. Multimedia Tools and Applications,2024,83(12):36329-36353.
[24] WANG Liwen,LIU Zhisong,SIU W C,et al. Lightening network for low-light image enhancement[J]. IEEE Transactions on Image Processing,2020,29:7984-7996. DOI: 10.1109/TIP.2020.3008396
[25] ZUIDERVELD K. Contrast limited adaptive histogram equalization[J]. Graphics Gems,1994,5:474-485.
[26] KIM Y,KOH Y,LEE C,et al. Dark image enhancement based on pairwise target contrast and multi-scale detail boosting[C]. IEEE International Conference on Image Processing,Quebec City,2015:1404-1408.
[27] WANG Shuhang,ZHENG Jin,HU Haimiao,et al. Naturalness preserved enhancement algorithm for non-uniform illumination images[J]. IEEE Transactions on Image Processing,2013,22(9):3538-3548. DOI: 10.1109/TIP.2013.2261309
[28] LI Mading,LIU Jiaying,YANG Wenhan,et al. Structure-revealing low-light image enhancement via robust Retinex model[J]. IEEE Transactions on Image Processing,2018,27(6):2828-2841. DOI: 10.1109/TIP.2018.2810539
计量
- 文章访问数: 126
- HTML全文浏览量: 30
- PDF下载量: 22
