A traffic image dehaze method based on adaptive transmittance estimation with multi-scale window
-
摘要: 基于传统暗原色先验原理的图像去雾算法存在的"halo"效应,且图像中明亮区域存在颜色失真现象,针对此问题,本文提出了多尺度窗口的自适应透射率修复交通图像去雾方法。首先,利用新的8方向边缘检测算子求取图像中景深突变区域,根据暗通道先验理论和前一步求得的景深突变区域,在景深变化较大区域使用5×5的窗口,景深变化较小区域则使用15×15的窗口得到暗原色估计图。同时,针对暗通道先验原理对近景部分存在白色区域时透射率估计不准确的问题,引入了自适应透射率修复方法,通过引导滤波器得到边缘增强后的暗原色图像,并利用其与原暗原色图像的纹理差对近景区域的透射率进行修正,完成图像去雾。实验结果表明:双边滤波和梯度双边滤波两种算法均存在halo现象,并且在包含白色物体的明亮区域色彩失真严重,客观评价指标失去意义;相比于引导滤波,本文去雾算法的各项指标均有所提高,其中平均梯度平均提高了8.305%,PSNR平均提高了12.455%,边缘强度因子平均提高了7.77%。本文算法有效解决了复原图像中"halo"效应现象和明亮区域颜色失真现象,去雾效果最优。Abstract: Aiming at the halo effect and the color distortion of bright areas when using traditional dark priori image defogging algorithms, we propose a traffic image dehaze method based on adaptive transmittance estimation with multi-scale window in this paper. Firstly, a new 8-direction edge detection operator is used to detect abrupt changes in field depth in images. According to the dark channel prior theory and the abrupt change of field depth obtained in the previous step, a 5×5 window is used in the larger area of field depth transformation and a 15×15 window is used in the smaller area to obtain a dark primary color estimation image. At the same time, targetting the problem of inaccurate estimation of transmittance when there is a white area in the close-range region due to the dark channel priori principle, we introduce an adaptive transmittance restoration method. An edge-enhanced dark image is obtained by using a guide filter, and the texture difference between the edge-enhanced dark image and the original dark primary image is used to correct the transmittance in the close-range region, and then to complete image dehazing. The experimental results show that the halo phenomenon exists in both the bilateral filter and the gradient bilateral filter, and the color distortion is serious in the bright area containing white objects, causing the objective evaluation index to be meaningless. Compared with the guide filter, the indexes of the dehazing algorithm used in this paper show improvement, wherein the average gradient increased by 8.305%, the PSNR increased by 12.455% and the edge strength factor increased by 7.77%. The algorithm can effectively solve issues arising from the halo effect and color distortion in bright areas in restored images while providing a more effective dehazing effect.
-
Key words:
- dark channel /
- image dehazing /
- traffic image /
- edge detection algorithm
-
表 1 两种算法的时间复杂度
Table 1. Consuming times of two kinds of algorithms
算法名称 时间(s) 插值抠图算法 10.732 多尺度算法 0.304 表 2 第一组实验结果的定量评价结果
Table 2. Evaluation indexes of the first set of experimental results
表 3 第二组实验结果的定量评价结果
Table 3. Evaluation indexes of the second set of experimental results
-
[1] SINGH D, KUMAR V. Single image haze removal using integrated dark and bright channel prior[J]. Modern Physics Letters B, 2018, 32(4):1850051. doi: 10.1142/S0217984918500513 [2] FATTAL R. Single image dehazing[J]. ACM Transactions on Graphics, 2008, 27(3):72. http://d.old.wanfangdata.com.cn/Periodical/jsjyjyfz201012013 [3] ZHANG Z, FENG W, WANG T, et al.. An improved aerial remote sensing image defogging method based on dark channel prior information[C]. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, ISPRS, 2017. [4] TAN K, OAKLEY J P. Enhancement of color images in poor visibility conditions[C]. Proceedings 2000 International Conference on Image Processing, IEEE, 2000: 788-791. [5] NARASIMHAN S G, NAYAR S K. Contrast restoration of weather degraded images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2003, 25(6):720-724. http://d.old.wanfangdata.com.cn/OAPaper/oai_doaj-articles_429f21ea83a55d2761e1aa45a29e1537 [6] LI Z G, ZHENG J H, YAO W, et al.. Single image haze removal via a simplified dark channel[C]. 2015 IEEE International Conference on Acoustics, Speech and Signal Processing, IEEE, 2015. [7] HE K M, SUN J, TANG X O. Single image haze removal using dark channel prior[C]. 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009: 1956-1963. [8] 黄大荣, 方周, 赵玲.一种改进的结合边缘检测的去雾新方法[J].上海交通大学学报, 2015, 49(6):861-867. http://d.old.wanfangdata.com.cn/Periodical/shjtdxxb201506021HUANG D R, FANG ZH, ZHAO L. An improved defogging algorithm combined with edge detection[J]. Journal of Shanghai Jiaotong University, 2015, 49(6):861-867.(in Chinese) http://d.old.wanfangdata.com.cn/Periodical/shjtdxxb201506021 [9] 石文轩, 詹诗萦, 李婕.一种边缘优化的暗通道去雾算法[J].计算机应用研究, 2013, 30(12):3854-3856, 3862. doi: 10.3969/j.issn.1001-3695.2013.12.088SHI W X, ZHAN SH Y, LI J. Dark channel prior dehazing algorithm based on edge optimization[J]. Application Research of Computers, 2013, 30(12):3854-3856, 3862.(in Chinese) doi: 10.3969/j.issn.1001-3695.2013.12.088 [10] 杨红, 崔艳.基于开运算暗通道和优化边界约束的图像去雾算法[J].光子学报, 2018, 47(6):238-244. http://d.old.wanfangdata.com.cn/Periodical/gzxb201806028YANG H, CUI Y. Image defogging algorithm based on opening dark channel and improved boundary constraint[J]. Acta Photonica Sinica, 2018, 47(6):238-244.(in Chinese) http://d.old.wanfangdata.com.cn/Periodical/gzxb201806028 [11] VERMA O P, PARIHAR A S. An optimal fuzzy system for edge detection in color images using bacterial foraging algorithm[J]. IEEE Transactions on Fuzzy Systems, 2017, 25(1):114-127. doi: 10.1109/TFUZZ.2016.2551289 [12] SIDOR K, SZLACHTA A. The impact of the implementation of edge detection methods on the accuracy of automatic voltage reading[J]. Measurement Science Review, 2017, 17(2):93-99. doi: 10.1515/msr-2017-0012 [13] LIU Y, CHENG M M, HU X W, et al.. Richer convolutional features for edge detection[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition, IEEE, 2017. [14] 汪贵平, 宋京, 杜晶晶, 等.基于改进梯度相似度核的交通图像去雾算法[J].中国公路学报, 2018, 31(6):264-271, 280. doi: 10.3969/j.issn.1001-7372.2018.06.013WANG G P, SONG J, DU J J, et al.. Haze defogging algorithm for traffic images based on improved gradient similarity kernel[J]. China Journal of Highway and Transport, 2018, 31(6):264-271, 280.(in Chinese) doi: 10.3969/j.issn.1001-7372.2018.06.013 [15] 黄鹤, 宋京, 郭璐, 等.基于新的中值引导滤波的交通视频去雾算法[J].西北工业大学学报, 2018, 36(3):414-419. doi: 10.3969/j.issn.1000-2758.2018.03.002HUANG H, SONG J, GUO L, et al.. A novel dehazing algorithm based on median guide filter for traffic video[J]. Journal of Northwestern Polytechnical University, 2018, 36(3):414-419.(in Chinese) doi: 10.3969/j.issn.1000-2758.2018.03.002 [16] 邓莉.针对明亮区域的自适应全局暗原色先验去雾[J].光学 精密工程, 2016, 24(4):892-901. http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201604026DENG L. Adaptive image dehazing for bright areas based on global dark channel prior[J]. Opt. Precision Eng., 2016, 24(4):892-901.(in Chinese) http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201604026 [17] 徐伟, 陈彦彤, 朴永杰, 等.基于吉林一号遥感图像的星载目标快速识别系统[J].光学 精密工程, 2017, 25(1):255-262. http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201701032XU W, CHEN Y T, PIAO Y J, et al.. Target fast matching recognition of on-board system based on Jilin-1 satellite image[J]. Opt. Precision Eng., 2017, 25(1):255-262. (in Chinese) http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201701032 [18] 李佳童, 章毓晋.图像去雾算法的改进和主客观性能评价[J].光学 精密工程, 2017, 25(3):735-741. http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201703025LI J T, ZHANG Y J. Improvements of image haze removal algorithm and its subjective and objective performance evaluation[J]. Opt. Precision Eng., 2017, 25(3):735-741.(in Chinese) http://d.old.wanfangdata.com.cn/Periodical/gxjmgc201703025