A Dehaze Algorithm Based on Near-Infrared and Visible Dual Channel Sensor Information Fusion
SHEN Yu1，2，3, DANG Jian-wu1，2*, GOU Ji-xiang4, GUO Rui1，2, LIU Cheng1，2, WANG Xiao-peng1, LI Lei1，2
1. School of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
2. Gansu Provincial Engineering Research Center for Artificial Intelligence and Graphics & Image Processing, Lanzhou 730070, China
3. Key Laboratory of Opto-technology and Intelligent Control, Ministry of Education, Lanzhou Jiaotong University, Lanzhou 730070, China
4. Troop 68003, PLA, Wuwei 733000, China
Abstract：In order to defog the image under hazy weather, multiple images defogging algorithm is one of the commonly used methods. Multiple images defogging algorithm also takes many forms, some of which are usually confronted with the problems of difficult hardware implementation, limited data source achievement approaches, or poor implementation et al. Meanwhile, multiple images defogging algorithm usually needs image registration in the comparison process, causing poor real-time performance and high computation cost. For the above problems, this study supplies a new idea for multiple images defogging algorithm, the near-infrared sensor images are used as new data source. The near-infrared sensor could penetrate haze to some extent, capturing the image details that the visible light sensor could not get. Meanwhile, the hardware of dual sensor system is simple. In the dual sensor system, the visible light image has abundant color information and the near-infrared sensor image can better describe the scene details at close range. The captured images could be completely registered with little rectification. The fusion of the infrared image and visible light image could extract the image details of the near-infrared image to the color visible light image to get the defogging image with abundant edge and contour information. Therefore, this study proposed a defogging algorithm using near-infrared and visible light image fusion method based on the edge details descriptive ability of the near-infrared sensor and the expressive ability of color information. Firstly, the color visible image was transformed from RGB space to HIS color space to get the hue channel image, saturation channel image and intensity channel image. The intensity channel image and original near-infrared image were decomposed by the Non-subsampled Shearlet Transform (NSST) method to get the high frequency coefficients and the low frequency coefficients. The high frequency component was treated by the double-exponential edge smoothing filter and the low frequency component was treated by the Unsharp Masking method, then the fusion rules and the inverse NSST were adopted to get the new intensity channel image. For the color information treatment of the visible light image, the degeneration model of the saturation channel image was established and the dark channel prior was used to evaluate its parameters to get the new saturation channel image. Finally, the new intensity channel image, the new saturation channel image and the original hue channel image were inversely mapped to the RGB space to get the defogging image. In order to verify the algorithm, we adopted 4 groups of foggy near-infrared images and visible light images as the experimental data. The processed images were compared with the defogging images observed by other two defogging algorithms. The experiment results showed that the proposed algorithm has better effect in improving the edge contrast and visual clarity. This study put forward the near-infrared image as new data source and the binary channels image fusion algorithm as the defogging method, and it was verified that the new algorithm for image defogging is feasible. This algorithm has four main advantages. The first one is that we combined the image fusion method with the defogging algorithm to get a novel idea for defogging algorithm. We transformed the color visible image to HSI color space. The obtained intensity channel image and original near-infrared image were fused by the NSST method, and the image details in the near-infrared image were simultaneously extracted to the color visible image in the defogging processes. The defogged image has abundant detailed information of edge and rough. The second advantage is that this algorithm adopted near-infrared sensor image as new data source. From the perspective of image processing, the near-infrared sensor could penetrate haze to some extent, capturing the image details that the visible light sensor could not get. Meanwhile, the hardware of dual sensor system is simple. The third advantage is that we adopted multiple images defogging algorithm, which captured images by binocular sensor system, and the visible light sensor got the image with abundant color information and the near-infrared sensor got the image with good detail description ability of close shot. The fusion of the two kinds of images has better effect than the single image defogging algorithm. The fourth advantage is that the visible light image was transformed to the HIS color space, and the images of the three channels can be targetedly processed according to their data characteristics. The process of intensity channel image of visible light image and the near-infrared image adopted image fusion and enhancement methods. The process of saturation channel image of visible light image adopted image restoration method. These processing enhances the effect of defogged image on the whole. This study supplies a new technological approach and way for image defogging algorithm.
沈 瑜，党建武，苟吉祥，郭 瑞，刘 成，王小鹏，李 磊. 近红外与可见光双通道传感器信息融合的去雾技术[J]. 光谱学与光谱分析, 2019, 39(05): 1420-1427.
SHEN Yu, DANG Jian-wu, GOU Ji-xiang, GUO Rui, LIU Cheng, WANG Xiao-peng, LI Lei. A Dehaze Algorithm Based on Near-Infrared and Visible Dual Channel Sensor Information Fusion. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2019, 39(05): 1420-1427.
[1] ZHANG Yong-yan, WU Jiu-hui, ZENG Tao, et al(张永燕, 吴九汇, 曾 涛, 等).Acta Physics Sinaca(物理学报), 2016，65：158.
[2] XIA Pu, LIU Xue-bin(夏 璞, 刘学斌). Spectroscopy and Spectral Analysis(光谱学与光谱分析)，2017，37(8)：2331.
[3] LIU Guo, Lü Qun-bo, LIU Yang-yang(刘 国, 吕群波, 刘扬阳). Acta Photonica Sinaca(光子学报), 2018，47：179.
[4] He K M, Sun J, Tang X O. IEEE Transactions on Pattern Analysis and Machine Intelligence，2011，33：2341.
[5] Miyazaki D, Akiyama D, Baba M, et al. Polarization-Based Dehazing Using Two Reference Objects. IEEE International Conference on Computer Vision Workshops，2014. 852.
[6] Yadav G, Maheshwari S, Agarwal A. Fog Removal Techniques from Images: A Comparative Review and Future Directions. International Conference on Signal Propagation and Computer Technology(ICSPCT)，2014. 44.
[7] Zhang T, Shao C，Wang X. Atomospheric Scattering-Based Multiple Images Fog Removal. 4th International Congress on Image and Signal Processing，2011，108.
[8] Jiang X F, Ma W. Research of New Method for Removal Thin Cloud and Fog of the Remote Sensing Images. Symposium on Photonics and Optoelectronics，SOPO 2010-Proc. 10.1109/SOPO. 2010. 5504180.
[9] LIU Tong-jun(刘同军). Journal of Hunan University of Technology(湖南工业大学学报)，2016，30：37.
[10] Li Nan, Lu Xiaobo. Journal of Southeast University(English Edition)，2011，27：290.
[11] WANG Jian-xin, ZHANG You-hui, WANG Zhi-wei(王建新, 张有会, 王志巍). Journal of Computer Applications(计算机应用)，2014，34：2990.
[12] Guo K, Labate D. Journal on Mathematical Analysis，2007，39：298.
[13] Philippe T, Daniel S, Unser M. IEEE Transactions on Image Processing，2012，21：3924.
[14] Tarel J P, Hautiere N. Fast Visibility Restoration from a Single Color or Gray Level Image. IEEE International Conference on Computer Vision(ICCV’09)，Washington, DC: IEEE Xplore, 2009. 2201.
