-
当人们在低照度光照条件下拍摄图像时, 图像通常会受到低可见度的影响. 这种低可见度的图像不仅影响视觉效果而且对后续的使用造成诸多困难. 为了解决低照度条件下图像可见度差, 色彩偏差等问题, 本文提出了一种改进的Retinex网络增强方法. 该方法首先对低照度RGB图像进行HSV色彩空间变换, 利用Retinex分解网络单独对明度分量进行分解增强, 并通过上采样操作增大明度分量的分辨率. 然后对色相分量和饱和度分量, 运用最近邻点插值增大其分辨率, 结合增强的明度分量转换回RGB色彩空间, 得到初始增强图像. 最后采用小波变换图像融合技术, 与原始低照度图像进行融合, 消除初始增强图像中的过度增强部分. 实验结果分析表明, 本文所提方法与原始Retinex网络方法相比, NIQE值平均下降了19.49%, 图像标准差平均提升了41.35%. 本文所提算法有望在安防监控、生物医学等领域得到有效应用.When capturing images under low-light lighting conditions, the resulting images often suffer low visibility. Such low-visibility images not only affect the visual effect but also cause many difficulties in practical application. Therefore, image enhancement technology under low-light conditions has always been a challenging problem in image algorithms. Considering that most of the existing image enhancement methods are based on the RGB color space enhancement technology, the correlation among the RGB three primary colors is ignored, which makes the color distortion phenomenon easy to occur when the image is enhanced. To solve the problems of poor image visibility and color deviation under low-light conditions, in this paper an advanced Retinex network enhancement method is proposed. In the method, firstly the low-light RGB image is transformed into HSV color space, the Retinex decomposition network is used to decompose and enhance the value component separately, and thus increasing the resolution of the value component through up-sampling operation; then, for the hue component and saturation component, the nearest neighbor interpolation is used to increase their resolutions, and the enhanced value component is combined to convert back to RGB color space to obtain the initial enhanced image; finally, the wavelet transform image fusion technology is used to fuse with the original low-light image to eliminate the over-enhanced part in the initial enhanced image. The analysis of experimental results shows that the method proposed in this paper has obvious advantages in brightness enhancement and color restoration of low-light images. Especially, comparing with the original Retinex network method, the NIQE value decreases by an average of 19.49%, and the image standard deviation increases by an average of 41.35%. The algorithm proposed in this paper is expected to be effectively used in the fields of security monitoring and biomedicine.
-
Keywords:
- image enhancement /
- Retinex /
- deep learning /
- image fusion
[1] 蒋一纯, 詹伟达, 朱德鹏 2021 激光与光电子学进展 58 0410001
Google Scholar
Jiang Y C, Zhan W D, Zhu D P 2021 Laser Optoelectron. Prog. 58 0410001
Google Scholar
[2] 韩平丽, 刘飞, 张广, 陶禹, 邵晓鹏 2018 67 054202
Google Scholar
Han P L, Liu F, Zhang G, Tao Y, Shao X P 2018 Acta Phys. Sin. 67 054202
Google Scholar
[3] Liu J, Wang X, Chen M, Liu S G, Zhou X R, Shao Z F, Liu P 2014 Opt. Express 22 618
Google Scholar
[4] Fu X Y, Zeng D L, Huang Y, Liao Y H, Ding X H, Paisley J 2016 Signal Process. 129 82
Google Scholar
[5] Singh N, Bhandari A K 2021 IEEE Trans. Instrum. Meas. 70 1
[6] Land E H 1964 Am. Sci. 52 247
[7] Land E H, McCann J J 1971 J. Opt. Soc. Am. 61 1
Google Scholar
[8] Land E H, Hubel D H, Livingstone M S, Perry S H, Burns M M 1983 Nature 303 616
Google Scholar
[9] 李红, 吴炜, 杨晓敏, 严斌宇, 刘凯, Gwanggil J 2016 65 160701
Google Scholar
Li H, Wu W, Yang X M, Yan B Y, Liu K, Gwanggil J 2016 Acta Phys. Sin. 65 160701
Google Scholar
[10] Jobson D J, Rahman Z, Woodell G A 1997 IEEE Trans. Image Process. 6 451
Google Scholar
[11] Rahman Z, Jobson D J, Woodell G A 1996 Proceedings of 3rd IEEE International Conference on Image Processing Lausanne, Switzerland, September 19, 1996 p1003
[12] Jobson D J, Rahman Z, Woodell G A 1997 IEEE Trans. Image Process. 6 965
Google Scholar
[13] 毕国玲, 续志军, 赵建, 孙强 2015 64 100701
Google Scholar
Bi G L, Xu Z J, Zhao J, Sun Q 2015 Acta Phys. Sin. 64 100701
Google Scholar
[14] Zhou Z Q, Dong M J, Xie X Z, Gao Z F 2016 Appl. Opt. 55 6480
[15] 王殿伟, 韩鹏飞, 范九 伦, 刘颖, 许志杰, 王晶 2018 67 210701
Google Scholar
Wang D W, Han P F, Fan J L, Liu Y, Xu Z J, Wang J 2018 Acta Phys. Sin. 67 210701
Google Scholar
[16] Kwon H J, Lee S H, Lee G Y, Sohng K I 2014 Digit. Signal Process. 30 74
Google Scholar
[17] Yang Q X, Tan K H, Ahuja N 2009 IEEE Conference on Computer Vision and Pattern Recognition Miami, USA, June 20–25, 2009 p557
[18] Wang S H, Zheng J, Hu H M, Li B 2013 IEEE Trans. Image Process. 22 3538
Google Scholar
[19] Fu X Y, Zeng D L, Huang Y, Zhang X P, Ding X H 2016 IEEE Conference on Computer Vision and Pattern Recognition Las Vegas, USA, June 27–30, 2016 p2782
[20] Guo X J, Li Y, Ling H B 2017 IEEE Trans. Image Process. 26 982
Google Scholar
[21] Gijsenij A, Gevers T, Weijer J 2011 IEEE Trans. Image Process. 20 2475
Google Scholar
[22] 赵欣慰, 金韬, 池灏, 曲嵩 2015 64 104201
Google Scholar
Zhao X W, Jin T, Chi H, Qu S 2015 Acta Phys. Sin. 64 104201
Google Scholar
[23] Jiang Z Q, Li H T, Liu L j, Men A D, Wang H Y 2021 Neurocomputing 454 361
Google Scholar
[24] 马红强, 马时平, 许悦雷, 朱明明 2019 光学学报 39 0210004
Google Scholar
Ma H Q, Ma S P, Xu Y L, Zhu M M 2019 Acta Opt. Sin. 39 0210004
Google Scholar
[25] Guo Y H, Ke X, Ma J, Zhang J 2019 IEEE Access 7 13737
Google Scholar
[26] Lore K G, Akintayo A, Sarkar S 2017 Pattern Recognit. 61 650
Google Scholar
[27] Wang W J, Wei C, Yang W H, Liu J Y 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition Xi'an, China, May 15–19, 2018 p751
[28] He W J, Liu Y Y, Feng J F, Zhang W W, Gu G H, Chen Q 2020 IEEE 3rd International Conference on Information Systems and Computer Aided Education Dalian, China, September 27–29, 2020 p397
[29] Wei C, Wang W J, Yang W H, Liu J Y 2018 arXiv: 1808.04560 v1 [cs. CV]
[30] Yakno M, Mohamad-Saleh J, Ibrahim M Z 2021 Sensors 21 6445
Google Scholar
[31] 陈刚, 刘言, 杨贺超, 孙斌, 喻春雨 2021 光学精密工程 29 1999
Google Scholar
Chen G, Liu Y, Yang H C, Sun B, Yu C Y 2021 Opt. Precis. Eng. 29 1999
Google Scholar
[32] Zhang H Y, Cao L C, Yang F 2021 Proc. SPIE First Optics Frontier Conference Hangzhou, China, June 18, 2021 1185002
[33] Yadav A K, Roy R, Kumar A P, Kumar C S, Dhakad S K 2015 International Conference on Advances in Computing, Communications and Informatics Kochi, India, August 10–13, 2015 p1204
[34] Mittal A, Soundararajan R, Bovik A C 2013 IEEE Signal Process. Lett. 20 209
Google Scholar
-
图 1 (a) Retinex成像理论模型; (b) Retinex算术思想简介
Fig. 1. (a) Retinex imaging theoretical model; (b) arithmetic ideas of Retinex algorithm.
图 2 改进的Retinex网络增强算法流程图
Fig. 2. Flow chart of the advanced Retinex network enhancement algorithm.
图 5 不同算法增强效果局部放大图
Fig. 5. Local enlarged view of the enhancement effect of different algorithms.
图 6 不同方法图像评价指标的均值变化情况
Fig. 6. Changes in mean values of image evaluation metrics for different methods.
表 1 V分量分解网络结构
Table 1. V-component decomposition network structure.
输入 操作 卷积核 输出通道 步长 输出 RGB rgb to hsv — — — H, S, V V conv 3 64 1 feats0 feats0 conv & ReLU 3 64 1 feats1 feats1 conv & ReLU 3 64 1 feats2 feats2 conv & ReLU 3 64 1 feats3 feats3 conv & ReLU 3 64 1 feats4 feats4 conv & ReLU 3 64 1 feats5 feats5 conv & sigmoid 3 2 1 R, I 
下载: 导出CSV
表 2 V分量增强网络结构
Table 2. V-component enhancement network structure.
输入 操作 卷积核 输出通道 步长 输出 Vlow, Rlow, Ilow up-sample — — — Input Input conv 3 64 1 out0 out0 conv & ReLU 3 64 2 out1 out1 conv & ReLU 3 64 2 out2 out2 conv & ReLU 3 64 2 out3 out3 interpolation — 64 — out3 up out3 up, out2
de1conv & ReLU
interpolation3
—64
641
—de1
de1 upde1 up, out1
de2conv & ReLU
interpolation3
—64
641
—de2
de2 upde2 up, out0de1
de2conv & ReLUinterpolation
interpolation3—
—6464
641—
—de3de1 r
de2 rde1 r, de2 r, de3 conv & ReLU 3 64 1 feats0 feats0 conv 1 64 1 feats1 feats1 conv 3 1 1 Vnew
下载: 导出CSV
表 3 不同图像的客观评价指标
Table 3. Objective evaluation metrics for different images.
Image Evaluate MSRCR Auto GC Retinex-Net ARN Image1 NIQE 5.6692 5.1384 5.9782 4.0729 Entropy 7.1095 6.6392 7.1375 7.8179 SD 33.3758 41.6959 31.1130 42.9601 Image 2 NIQE 6.2926 6.0252 5.3596 3.7336 Entropy 7.3012 7.5171 7.5777 7.7226 SD 41.8903 55.6071 46.3428 66.2911 Image 3 NIQE 5.6715 4.9203 4.4528 4.0319 Entropy 6.7898 7.1224 7.7284 7.8633 SD 31.3800 37.3028 53.5654 72.4424 Image 4 NIQE 3.7695 3.8844 3.7200 3.6582 Entropy 5.5392 7.1881 7.2807 7.4010 SD 40.8917 38.6741 39.9913 46.9674 Image 5 NIQE 3.9541 4.4738 4.0126 3.6424 Entropy 7.3497 6.0549 7.2871 7.4387 SD 42.1574 41.0863 32.5321 56.5474 Image 6 NIQE 7.3401 6.4273 5.4459 3.8790 Entropy 7.0335 5.5701 7.3417 7.8134 SD 34.1136 40.6108 38.4974 56.5800 Mean NIQE 5.4495 5.1449 4.7649 3.8363 Entropy 6.8538 6.6820 7.3922 7.6762 SD 37.3015 42.4962 40.3003 56.9647
下载: 导出CSV
-
[1] 蒋一纯, 詹伟达, 朱德鹏 2021 激光与光电子学进展 58 0410001
Google Scholar
Jiang Y C, Zhan W D, Zhu D P 2021 Laser Optoelectron. Prog. 58 0410001
Google Scholar
[2] 韩平丽, 刘飞, 张广, 陶禹, 邵晓鹏 2018 67 054202
Google Scholar
Han P L, Liu F, Zhang G, Tao Y, Shao X P 2018 Acta Phys. Sin. 67 054202
Google Scholar
[3] Liu J, Wang X, Chen M, Liu S G, Zhou X R, Shao Z F, Liu P 2014 Opt. Express 22 618
Google Scholar
[4] Fu X Y, Zeng D L, Huang Y, Liao Y H, Ding X H, Paisley J 2016 Signal Process. 129 82
Google Scholar
[5] Singh N, Bhandari A K 2021 IEEE Trans. Instrum. Meas. 70 1
[6] Land E H 1964 Am. Sci. 52 247
[7] Land E H, McCann J J 1971 J. Opt. Soc. Am. 61 1
Google Scholar
[8] Land E H, Hubel D H, Livingstone M S, Perry S H, Burns M M 1983 Nature 303 616
Google Scholar
[9] 李红, 吴炜, 杨晓敏, 严斌宇, 刘凯, Gwanggil J 2016 65 160701
Google Scholar
Li H, Wu W, Yang X M, Yan B Y, Liu K, Gwanggil J 2016 Acta Phys. Sin. 65 160701
Google Scholar
[10] Jobson D J, Rahman Z, Woodell G A 1997 IEEE Trans. Image Process. 6 451
Google Scholar
[11] Rahman Z, Jobson D J, Woodell G A 1996 Proceedings of 3rd IEEE International Conference on Image Processing Lausanne, Switzerland, September 19, 1996 p1003
[12] Jobson D J, Rahman Z, Woodell G A 1997 IEEE Trans. Image Process. 6 965
Google Scholar
[13] 毕国玲, 续志军, 赵建, 孙强 2015 64 100701
Google Scholar
Bi G L, Xu Z J, Zhao J, Sun Q 2015 Acta Phys. Sin. 64 100701
Google Scholar
[14] Zhou Z Q, Dong M J, Xie X Z, Gao Z F 2016 Appl. Opt. 55 6480
[15] 王殿伟, 韩鹏飞, 范九 伦, 刘颖, 许志杰, 王晶 2018 67 210701
Google Scholar
Wang D W, Han P F, Fan J L, Liu Y, Xu Z J, Wang J 2018 Acta Phys. Sin. 67 210701
Google Scholar
[16] Kwon H J, Lee S H, Lee G Y, Sohng K I 2014 Digit. Signal Process. 30 74
Google Scholar
[17] Yang Q X, Tan K H, Ahuja N 2009 IEEE Conference on Computer Vision and Pattern Recognition Miami, USA, June 20–25, 2009 p557
[18] Wang S H, Zheng J, Hu H M, Li B 2013 IEEE Trans. Image Process. 22 3538
Google Scholar
[19] Fu X Y, Zeng D L, Huang Y, Zhang X P, Ding X H 2016 IEEE Conference on Computer Vision and Pattern Recognition Las Vegas, USA, June 27–30, 2016 p2782
[20] Guo X J, Li Y, Ling H B 2017 IEEE Trans. Image Process. 26 982
Google Scholar
[21] Gijsenij A, Gevers T, Weijer J 2011 IEEE Trans. Image Process. 20 2475
Google Scholar
[22] 赵欣慰, 金韬, 池灏, 曲嵩 2015 64 104201
Google Scholar
Zhao X W, Jin T, Chi H, Qu S 2015 Acta Phys. Sin. 64 104201
Google Scholar
[23] Jiang Z Q, Li H T, Liu L j, Men A D, Wang H Y 2021 Neurocomputing 454 361
Google Scholar
[24] 马红强, 马时平, 许悦雷, 朱明明 2019 光学学报 39 0210004
Google Scholar
Ma H Q, Ma S P, Xu Y L, Zhu M M 2019 Acta Opt. Sin. 39 0210004
Google Scholar
[25] Guo Y H, Ke X, Ma J, Zhang J 2019 IEEE Access 7 13737
Google Scholar
[26] Lore K G, Akintayo A, Sarkar S 2017 Pattern Recognit. 61 650
Google Scholar
[27] Wang W J, Wei C, Yang W H, Liu J Y 2018 13th IEEE International Conference on Automatic Face & Gesture Recognition Xi'an, China, May 15–19, 2018 p751
[28] He W J, Liu Y Y, Feng J F, Zhang W W, Gu G H, Chen Q 2020 IEEE 3rd International Conference on Information Systems and Computer Aided Education Dalian, China, September 27–29, 2020 p397
[29] Wei C, Wang W J, Yang W H, Liu J Y 2018 arXiv: 1808.04560 v1 [cs. CV]
[30] Yakno M, Mohamad-Saleh J, Ibrahim M Z 2021 Sensors 21 6445
Google Scholar
[31] 陈刚, 刘言, 杨贺超, 孙斌, 喻春雨 2021 光学精密工程 29 1999
Google Scholar
Chen G, Liu Y, Yang H C, Sun B, Yu C Y 2021 Opt. Precis. Eng. 29 1999
Google Scholar
[32] Zhang H Y, Cao L C, Yang F 2021 Proc. SPIE First Optics Frontier Conference Hangzhou, China, June 18, 2021 1185002
[33] Yadav A K, Roy R, Kumar A P, Kumar C S, Dhakad S K 2015 International Conference on Advances in Computing, Communications and Informatics Kochi, India, August 10–13, 2015 p1204
[34] Mittal A, Soundararajan R, Bovik A C 2013 IEEE Signal Process. Lett. 20 209
Google Scholar
下载:
计量
- 文章访问数: 6469
- PDF下载量: 145
- 被引次数: 0
