Low-light Image Enhancement Based on Multi-scale Wavelet U-Net
-
摘要: 成像系统实时采集的低光照环境图像具有照度低、噪声严重、视觉效果差等问题,为了提高低光照环境成像质量,本文提出基于多尺度小波U型网络的低光照图像增强方法。该方法采用多级编解码器构建U型网络,并引入小波变换构建特征分频单元,分离高频和低频信息,增强对低频照度特征和高频纹理信息的感知。设计多尺度感知损失函数,指导网络学习低频信息到高频信息的逐级重建,从而优化网络的收敛和性能。最后,在LOL、LIME、NPE、MEF、DICM和VV数据集上进行测试。实验结果表明,所提方法能够有效提升图像照度,抑制图像噪声和纹理丢失问题,并在PSNR、SSIM、LOE和NIQE评价指标上均优于比较算法,在主观和客观评价方面均优于其他对比算法。Abstract: The real-time images collected by imaging systems in low illumination environments suffer from low illumination, severe noise and poor visual effects. To improve the image quality in low light environments, this study proposes a low-light image enhancement method based on a multi-scale wavelet U-Net. This method uses multi-level encoders and decoders to construct a U-Net and introduces the wavelet transform to develop a unit for frequency decomposition of features. This improves the perception of illumination and texture by separating high-frequency and low-frequency information. Multi-scale perceived loss is designed to guide the learned mapping of step-by-step reconstruction from low-frequency information to high-frequency information, thereby optimizing the convergence and performance of the network. Finally, the proposed method and comparison methods are tested on the LOL, LIME, NPE, MEF, DICM and VV datasets. The experimental results demonstrate that the proposed method can effectively brighten images, suppress image noise and texture loss, and improve the PSNR, SSIM, LOE and NIQE metrics. The proposed method exhibits better performance than other comparison algorithms in subjective and objective evaluation.
-
Key words:
- imaging systems /
- image enhancement /
- U-Net /
- wavelet transform
-
表 1 不同低光照增强算法在LOL数据集上的结果
Table 1. Results of different low light enhancement algorithms on LOL dataset
Metrics Low BPDHE LIME MF Retinexnet BIMEF KinD Ours PSNR 7.8666 11.8748 16.9332 16.9332 16.9816 13.9722 20.5955 24.5499 SSIM 0.1259 0.3415 0.5644 0.6048 0.5594 0.5771 0.8045 0.8913 LOE 1417.1 1772.8 3207.6 1721.2 4244.1 1553.3 1795.3 1267.1 NIQE 6.7483 7.9852 8.3777 8.8855 8.8779 7.5029 5.3561 5.4289 表 2 不同网络的编/解码器和损失函数的配置
Table 2. Configuration of codecs and loss function for different networks
Combination Single-scale L1 Multi-scale L1 Cascaded residual blocks FDF N1 √ - √ - N2 √ - - √ N3 - √ √ - N4 - √ - √ 表 3 不同组合网络的性能比较
Table 3. Performance comparison of different combination networks
Datasets LOL LIME NPE MEF DICM VV Average N1 6.8072 5.0207 6.4623 6.3248 6.5652 6.1562 6.2227 N2 4.8663 4.3757 5.3071 5.4867 5.0311 4.9305 4.9996 N3 4.8091 4.2395 6.2099 6.6436 6.0698 5.8492 5.6369 N4 4.6112 4.2251 4.9695 5.4682 4.9888 4.7840 4.8411 -
[1] 王萌萌, 彭敦陆. Retinex-ADNet: 一个低光照图像增强系统[J]. 小型微型计算机系统, 2022, 43(2): 367-371 https://www.cnki.com.cn/Article/CJFDTOTAL-XXWX202202021.htmWANG M M, PENG D L. Retinex-ADNet: a low-light image enhancement system[J]. Journal of Chinese Computer Systems, 2022, 43(2): 367-371. https://www.cnki.com.cn/Article/CJFDTOTAL-XXWX202202021.htm [2] 蒋一纯, 詹伟达, 朱德鹏. 基于亮度通道细节增强的低照度图像处理[J]. 激光与光电子学进展, 2021, 58(4): 83-91. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202104011.htmJIANG Y C, ZHAN W D, ZHU D P. Low-illuminance image processing based on brightness channel detail enhancement[J]. Laser & Optoelectronics Progress, 2021, 58(4): 83-91. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202104011.htm [3] 黄慧, 董林鹭, 刘小芳, 等. 改进Retinex的低光照图像增强[J]. 光学精密工程, 2020, 28(8): 1835-1849. https://www.cnki.com.cn/Article/CJFDTOTAL-GXJM202008022.htmHUANG H, DONG L L, LIU X F, et al. Improved Retinex low light image enhancement method[J]. Optics and Precision Engineering, 2020, 28(8): 1835-1849. https://www.cnki.com.cn/Article/CJFDTOTAL-GXJM202008022.htm [4] Ibrahim H, Pik Kong N S. Brightness preserving dynamic histogram equalization for image contrast enhancement[J]. IEEE Transactions on Consumer Electronics, 2007, 53(4): 1752-1758. doi: 10.1109/TCE.2007.4429280 [5] 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): 451-62. doi: 10.1109/83.557356 [6] Jobson D J, Rahman Z, Woodell G A. 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 [7] GAO Y, HU H M, LI B, et al. Naturalness preserved nonuniform illumination estimation for image enhancement based on Retinex[J]. IEEE Transactions on Multimedia, 2018, 20(2): 335-344. doi: 10.1109/TMM.2017.2740025 [8] FU X Y, ZENG D L, HUANG Y, et al. A fusion-based enhancing method for weakly illuminated images[J]. Signal Processing, 2016, 129: 82-96. doi: 10.1016/j.sigpro.2016.05.031 [9] GUO X J, LI Y, LING H B. LIME: Low-light image enhancement via illumination map estimation[J]. IEEE Transactions on Image Processing, 2017, 26(2): 982-993. doi: 10.1109/TIP.2016.2639450 [10] Kin G L, Adedotun A, Soumik S. LLNet: A deep autoencoder approach to natural low-light image enhancement[J]. Pattern Recognition, 2017, 61: 650-662. doi: 10.1016/j.patcog.2016.06.008 [11] SHEN L, YUE Z, FENG F, et al. MSR-net: low-light image enhancement using deep convolutional network[EB/OL]. [2017-11-07]. https://arxiv.org/abs/1711.02488. [12] WEI C, WANG W, YANG W, et al. Deep Retinex decomposition for low-light enhancement[C]//Proceedings of 2018 British Machine Vision Conference on Machine Vision, Image Processing, and Pattern Recognition, 2018: 1-12(DOI: 10.48550/ARXIV.1808.04560). [13] HU J, SHEN L, Albanie S, SUN G, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(8): 2011-2023. doi: 10.1109/TPAMI.2019.2913372 [14] SHI W, Caballero J, Huszár F, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 1874-1883. [15] Lim B, Son S, Kim H, et al. Enhanced deep residual networks for single image super-resolution[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops(CVPRW), 2017: 1132-1140. [16] Ledig C, Theis L, Huszar F, et al. Photo-realistic single image super-resolution using a generative adversarial network[C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition, 2017: 105-114. [17] MA K, ZENG K, WANG Z. Perceptual quality assessment for multi-exposure image fusion[J]. IEEE Transactions on Image Processing, 2015, 24(11): 3345-3356. doi: 10.1109/TIP.2015.2442920 [18] LEE C W, LEE C, KIM C S. Contrast enhancement based on layered difference representation of 2D histograms[J]. IEEE Transactions on Image Processing, 2013, 22(12): 5372-5384. doi: 10.1109/TIP.2013.2284059 [19] WANG S, ZHENG J, HU H M, 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 [20] Mittal A, Soundararajan R, Bovik A C. Making a "completely blind" image quality analyzer[J]. IEEE Signal Processing Letters, 2013, 20(3): 209-212. doi: 10.1109/LSP.2012.2227726 [21] FU X, ZENG D, HUANG Y, et al. A fusion-based enhancing method for weakly illuminated images[J]. Signal Processing, 2016, 129: 82-96. [22] YING Z, LI G, GAO W. A bio-inspired multi-exposure fusion framework for low-light image enhancement[EB/OL]. [2017-11-02]. https://arxiv.org/abs/1711.00591. [23] ZHANG Y, ZHANG J, GUO X. Kindling the darkness: a practical low-light image enhancer[C]//Proceedings of The 27th ACM International Conference on Multimedia, 2019: 1632-1640.