WANG Rong， YANG Yan
（School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China）
Abstract： Aiming at the problems such as color cast and incomplete haze removal in dehazing algorithms, a multi-level feature fusion network based on the learning of hazy layers is proposed for single image dehazing. Firstly, a difference image between hazy image and haze-free image is defined as the hazy layer via atmospheric scattering model, and the effective estimation of the hazy layer could be used to optimize dehazing effect. Then, an end-to-end network model is designed, which mainly includes a hazy layer estimation module and an image restoration module. In hazy layer estimation module, the low-level and high-level features of image are extracted through feature extraction blocks, and a multi-level fusion strategy is used to add the features of different levels pixel by pixel to achieve feature fusion. The fused hazy layer contains both local and global information. Finally, the hazy layer is directly subtracted from hazy image to achieve the effective restoration of haze-free image according to image restoration module. Experiments show that the proposed algorithm can obtain clear and natural results, and the color cast phenomenon is effectively avoided compared with existing dehazing methods. The objective evaluation indicators on synthetic images and real images further verify the effectiveness of the proposed algorithm.
Key words： image dehazing; convolutional neural network; hazy layer; feature extraction; multi-level fusion strategy
References
［1］WANG Y B, ZHENG J, YIN S B. Image dehazing method using channel attention network and fuzzy partition entropy with graph cut. Acta Photonica Sinica, 2021, 50(3): 310002.
［2］TAN R T. Visibility in bad weather from a single image// IEEE Conference on Computer Vision and Pattern Recognition. Jun.23-28, 2008, Anchorage, AK, USA. Anchorage: IEEE Press, 2008: 1-8.
［3］FATTAL R. Single image dehazing. ACM Transactions on Graphics, 2008, 27(3): 1-9.
［4］HE K M, SUN J, TANG X O. Single image haze removal using dark channel prior. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(12): 2341-2353.
［5］HE K M, SUN J, TANG X O. Guided image filtering. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013, 35(6): 1397-1409.
［6］MENG G F, WANG Y, DUAN J Y, et al. Efficient image dehazing with boundary constraint and contextual regularization//IEEE International Conference on Computer Vision. Dec. 8-15, 2013, Sydney, Australia. Sydney: IEEE, 2013: 617-624.
［7］ZHU Q S, MAI J M, SHAO L. A fast single image haze removal algorithm using color attenuation prior. IEEE Transactions on Image Processing, 2015, 24(11): 3522-3533.
［8］CAI B L, XU X M, JIA K, et al. DehazeNet: An end-to-end system for single image haze removal. IEEE Transactions on Image Processing, 2016, 25(11): 5187-5198.
［9］REN W Q, LIU S, ZHANG H, et al. Single image dehazing via multi-scale convolutional neural networks//European Conference on Computer Vision. Oct. 11-14, 2016, Amsterdam, Netherlands. Berlin: Springer International Publishing, 2016: 154-169.
［10］LI B Y, PENG X L, WANG Z Y, et al. AOD-Net: all-in-one dehazing network//IEEE International Conference on Computer Vision. Oct. 22-29, 2017, Venice, Italy. Venice: IEEE, 2017: 4780-4788.
［11］ZHANG H, PATEL V M. Densely connected pyramid dehazing network//IEEE Conference on Computer Vision and Pattern Recognition. Jun. 18-22, 2018, Salt Lake City, Utah, USA. New York: IEEE, 2018: 3194-3203.
［12］QU Y Y, CHEN Y Z, HUANG J Y, et al. Enhanced pix2pix dehazing network//IEEE Conference on Computer Vision and Pattern Recognition. Jun. 16-19, 2019, Los Angeles, USA. New York: IEEE, 2019: 8160-8168.
［13］QIN X, WANG Z L, BAI Y C, et al. FFA-Net: Feature fusion attention network for single image dehazing//Association for the Advance of Artificial Intelligence. Feb. 7-12, 2020, New York, USA. New York: AAAI, 2020: 11908-11915.
［14］NARASIMHAN S G, NAYAR S K. Vision and the Atmosphere. International Journal of Computer Vision, 2002, 48(3): 233-254.
［15］YANG Y, ZHANG C, JIANG P P, et al. Attention-based end-to-end image defogging network. Electronics Letters, 2020, 56(11): 759-761.
［16］HUANG G, LIU Z, LAURENS V, et al. Densely connected convolutional networks//IEEE Conference on Computer Vision and Pattern Recognition. Jul. 21-26, 2017, Hawaii, USA. New York: IEEE, 2017: 2261-2269.
［17］SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition. Computer Science, 2015, arXiv: 1409.1556.
［18］LI B Y, REN W Q, FU D P, et al. Benchmarking single image dehazing and beyond. IEEE Transactions on Image Processing, 2019, 28(1): 492-505.
［19］CHEN Y, GUO H G, AI Y P. Single image dehazing method based on multi-scale convolution neural network. Acta Optica Sinica, 2019, 39(10): 1010001.
［20］GUO F, CAI Z X. Objective assessment method for the clearness effect of image defogging algorithm. Acta Automatica Sinica, 2012, 38(9): 1410-1419.
基于雾层学习的多级融合去雾网络
王蓉， 杨燕
（兰州交通大学 电子与信息工程学院， 甘肃 兰州 730070）
摘要:针对去雾算法中存在的颜色失真、 去雾不彻底等问题， 本文提出一种基于雾层学习的多级特征融合网络用于单幅图像去雾。 首先， 结合大气散射模型将有雾图像与无雾图像之间的差值图像定义为雾度层， 通过对雾度层的有效估计来达到优化的去雾效果。 其次， 设计一种端到端的网络模型， 该模型主要包括雾层估计模块和图像复原模块。 在雾层估计模块中， 通过特征提取块对图像的低级和高级特征进行提取， 并采用多级融合策略对不同级别的特征进行逐像素加法来实现特征融合， 融合后的雾层特征同时包含局部信息和全局信息。 最后， 根据图像复原模块， 直接从有雾图像中减去雾层特征， 便可实现无雾图像的有效复原。 实验表明， 与现有去雾算法相比， 该算法复原结果清晰自然， 有效避免了颜色失真现象； 合成图像和真实图像上的客观评价指标进一步验证了所提算法的有效性。
关键词:图像去雾； 卷积神经网络； 雾度层； 特征提取； 多级融合策略
引用格式：WANG Rong， YANG Yan． Multi-level fusion dehazing network based on learning of hazy layers． Journal of Measurement Science and Instrumentation， 2023， 14（2）： 200-208. DOI： 10．3969／j．issn．1674-8042．2023．02．009
[full text view]
