密集连接扩张卷积神经网络的单幅图像去雾

doi:10.3778/j.issn.1673-9418.2001023

摘要/Abstract

摘要：

针对大多数图像去雾算法模型参数估计准确性差及色彩失真等问题，提出了一种端到端的密集连接扩张卷积神经网络。首先，通过使用多层密集连接结构来增加网络的特征利用率，避免网络加深时的梯度消失现象。其次，通过在密集块中使用不同扩张率的扩张卷积，使网络在充分聚合上下文特征信息时不损失空间分辨率，并避免了网格伪影的产生。最后，为了提高算法的去雾能力，将该网络划分为多个阶段，并在每个阶段引入侧输出模块，从而获得更精确的特征信息。实验结果表明，所提出的去雾算法无论是在合成数据集上还是在真实数据集上都取得了较好的去雾效果，恢复的色彩更接近无雾图像，并且定量评价指标峰值信噪比（PSNR）和结构相似性（SSIM）均优于其他对比方法。

关键词: 图像去雾, 卷积神经网络（CNN）, 密集连接, 扩张卷积

Abstract:

In view of the poor accuracy of model parameter estimation and color distortion in most image dehazing algorithms, an end-to-end dense connected dilated convolutional neural network is proposed. First of all, this paper uses multi-layer dense connection structure to increase the feature utilization of the network and avoid the gradient disappearance when the network deepens. Secondly, by using the dilation convolution with different dilation rates in the dense blocks, the network can fully aggregate the context feature information without losing the spatial resolution, and avoid the generation of mesh artifacts. Finally, in order to improve the ability of the algorithm, this paper divides the network into several stages, and introduces the side output module in each stage, so as to obtain more accurate feature information. The experimental results show that the proposed dehazing algorithm has a good dehazing effect on both the synthetic data set and the real data set. The recovered color is closer to the ground truth, and the quantitative evaluation indexes peak signal to noise ratio (PSNR) and structural similarity index (SSIM) are better than other comparison methods.

Key words: image dehazing, convolutional neural network (CNN), dense connection, dilation convolution

刘广洲, 李金宝, 任东东, 舒明雷. 密集连接扩张卷积神经网络的单幅图像去雾[J]. 计算机科学与探索, 2021, 15(1): 185-194.

LIU Guangzhou, LI Jinbao, REN Dongdong, SHU Minglei. Single Image Dehazing Method Based on Densely Connected Dilated Convolutional Neural Network[J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(1): 185-194.

参考文献

[1] MCCARTNEY E J. Optics of the atmosphere: scattering by molecules and particles[J]. Physics Today, 1976, 10(2): 461-470.
[2] TAN R T. Visibility in bad weather from a single image[C]// Proceedings of the 2008 IEEE Conference on Computer Vision and Pattern Recognition, Anchorage, Jun 24-26, 2008. Wash-ington: IEEE Computer Society, 2008: 1-8.
[3] HE K, SUN J, TANG X. Single image haze removal using dark channel prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 33(12): 2341-2353.
[4] TANG K T, YANG J C, WANG J. Investigating haze-relevant features in a learning framework for image dehazing[C]//Pro-ceedings of the 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, Jun 23-28, 2014. Wash-ington: IEEE Computer Society, 2014: 2995-3002.
[5] BERMAN D, TREIBITZ T, AVIDAN S. Non-local image deha-zing[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, Jun 27-30, 2016. Washington: IEEE Computer Society, 2016: 1674-1682.
[6] CAI B, XU X, JIA K, et al. Dehazenet: an end-to-end system for single image haze removal[J]. IEEE Transactions on Image Processing, 2016, 25(11): 5187-5198.
[7] REN W Q, LIU S, ZHANG H, et al. Single image dehazing via multi-scale convolutional neural neworks[C]//LNCS 9906: Proceedings of the 2016 European Conference on Computer Vision, Amsterdam, Oct 8-10, 2016. Berlin, Heidelberg: Spr-inger, 2016: 154-169.
[8] ZHANG H, PATEL V M. Densely connected pyramid dehazing network[C]//Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, Jun 18-22, 2018. Washington: IEEE Computer Society, 2018: 3194-3203.
[9] YANG D, SUN J. Proximal dehaze-net: a prior learning-based deep network for single image dehazing[C]//LNCS 11211: Proceedings of the 15th European Conference on Computer Vision, Munich, Sep 8-14, 2018. Berlin, Heidelberg: Springer, 2018: 729-746.
[10] LI R D, PAN J S, LI Z C, et al. Single image dehazing via conditional generative adversarial network[C]//Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, Jun 18-22, 2018. Washington: IEEE Computer Society, 2018: 8202-8211.
[11] REN W Q, MA L, ZHANG J W, et al. Gated fusion network for single image dehazing[C]//Proceedings of the 2018 IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, Jun 18-22, 2018. Washington: IEEE Computer Society, 2018: 3253-3261.
[12] LIU X H, MA Y R, SHI Z H, et al. GridDehazeNet: attention-based multi-scale network for image dehazing[C]//Proceed-ings of the 2019 IEEE International Conference on Computer Vision, Seoul, Oct 27-Nov 2, 2019. Piscataway: IEEE, 2019: 7314-7323.
[13] QU Y Y, CHEN Y Z, HUANG J Y, et al. Enhanced Pix2pix dehazing network[C]//Proceedings of the 2019 IEEE Con-ference on Computer Vision and Pattern Recognition, Long Beach, Jun 16-20, 2019. Piscataway: IEEE, 2019: 8160-8168.
[14] DENG Z J, ZHU L, HU X W, et al. Deep multi-model fusion for single-image dehazing[C]//Proceedings of the 2019 IEEE International Conference on Computer Vision, Seoul, Oct 27-Nov 2, 2019. Piscataway: IEEE, 2019: 2453-2462.
[15] GLOROT X, BORDES A, BENGIO Y. Deep sparse rectifier neural networks[C]//Proceedings of the 14th International Con-ference on Artificial Intelligence and Statistics, Fort Lauderdale, Apr 11-13, 2011. Cambridge: MIT Press, 2011: 315-323.
[16] HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recogni-tion, Honolulu, Jul 21-26, 2017. Washington: IEEE Computer Society, 2017: 2261-2269.
[17] YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions[J]. arXiv:1511.07122, 2015.
[18] JOHNSON J, ALAHI A, LI F F. Perceptual losses for real-time style transfer and super-resolution[C]//LNCS 9906: Procee-dings of the 2016 European Conference on Computer Vision, Amsterdam, Oct 11-14, 2016. Berlin, Heidelberg: Springer, 2016: 694-711.
[19] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. arXiv:1409.1556, 2014.
[20] RUSSAKOVSKY O, DENG J, SU H, et al. Imagenet large scale visual recognition challenge[J]. International Journal of Computer Vision, 2015, 115(3): 211-252.
[21] WANG Z, BOVIK A C, SHEIKH H R, et al. Image quality assessment: from error visibility to structural similarity[J]. IEEE Transactions on Image Processing, 2004, 13(4): 600-612.
[22] LI B, REN W, FU D, et al. Benchmarking single-image deha-zing and beyond[J]. IEEE Transactions on Image Processing, 2018, 28(1): 492-505.
[23] KINGMA D P, BA J. Adam: a method for stochastic optimiza-tion[J]. arXiv:1412.6980, 2014.
[24] LI B Y, PENG X L, WANG Z Y, et al. AOD-Net: all-in-one dehazing network[C]//Proceedings of the 2017 IEEE Interna-tional Conference on Computer Vision, Venice, Oct 22-29, 2017. Washington: IEEE Computer Society, 2017: 4770-4778.