Research on Low Illumination Image Enhancement Method Based on Spectral Reflectance
MA Xiang-cai1, 2, CAO Qian2, BAI Chun-yan2, WANG Xiao-hong3, ZHANG Da-wei1*
1. School of Optical Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
2. Department of Printing and Packaging Engineering, Shanghai Publishing and Printing College, Shanghai 20093, China
3. College of Communication and Art Design, University of Shanghai for Science and Technology, Shanghai 200093, China
Abstract：Low illumination image enhancement technology is one of the research hotspots of computer vision. The theoretical algorithm of Retinex assumes that the image is the product of the reflection component and the illumination component. It restores the image by removing or correcting the illumination component and combining the reflection component of the object, which is widely used in traditional algorithms and deep learning enhancement models. Spectral reflectance is the fingerprint of color, and multispectral images have more information than RGB images. Colorimetric theory and Retinex theory agree that the color of an image depends on reflection data, but spectral reflectance is obtained based on instrument measurement, and the image reflection component is obtained based on image hypothesis decomposition. The literature has not studied the enhancement of low-light images from the perspective of spectral reflectance. Inspired by Retinex theory and combined with the strong nonlinear fitting ability of deep learning, a low illumination image enhancement method based on spectral reflectance is proposed. The spectral reflectance of color is used to replace the image reflection component in the RetinexNet network, and the spectral power distribution of the CIE standard light source is used to replace the image illumination component in the network. Firstly, the spectral reflectance of normal light images in the image database is reconstructed to build a multispectral image dataset of low illumination and normal light images. Then, the deep learning network model is trained to convert low-illumination images into the multispectral images. Any low illuminance image is obtained from the multispectral image through the network model, and the multispectral image is obtained from the CIEXYZ tristimulus according to the colorimetric theory and then converted to the RGB color space for display through the standard color space.The method is trained and tested on the public LOL dataset, and the results show that this method is superior to the standard methods in image noise suppression and color restoration, which proves the superiority and effectiveness of this method for low illumination image enhancement.
麻祥才，曹 前，白春燕，王晓红，张大伟. 基于光谱反射率的低照度图像增强方法研究[J]. 光谱学与光谱分析, 2024, 44(03): 610-616.
MA Xiang-cai, CAO Qian, BAI Chun-yan, WANG Xiao-hong, ZHANG Da-wei. Research on Low Illumination Image Enhancement Method Based on Spectral Reflectance. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2024, 44(03): 610-616.
[1] Jobson D J, Rahman Z U, Woodel G A. IEEE Transactions on Image Processing, 1997, 6(3): 451.
[2] Jobson D J, Rahman Z, Woodell G A. IEEE Transactions on Image Processing, 1997, 6(7): 965.
[3] Rahman Z-U, Jobson D J, Woodell G A. Journal of Electronic Imaging, 2004, 13(1): 100.
[4] Park S, Yu S, Moon B, et al. IEEE Transactions on Consumer Electronics, 2017, 63(2): 178.
[5] Guo X, Li Y, Ling H. IEEE Transactions on Image Processing, 2017, 26(2): 982.
[6] Li C, Guo J, Porikli F, et al. Pattern Recognition Letters, 2018, 104: 15.
[7] Lü X, Sun Y, Zhang J, et al. Signal Processing: Image Communication, 2021, 99: 116466.
[8] Zhang Y, Zhang J, Guo X. Proceedings of the 27th ACM International Conference on Multimedia, 2019: 1632.
[9] Zhang Y, Di X, Zhang B, et al. arXiv: 2002. 11300v1, 2020.
[10] Liang J X, Xiao K D, Hu X R. Optics Express, 2021, 29(26): 43899.
[11] Wu G. Laser Physics Letters, 2019, 16 (8): 085208.
[12] XIAO Ying, MA Xiang-cai, WANG Dong-dong, et al(肖 颖, 麻祥才, 王东东, 等). Optical Technique(光学技术), 2019, 45(6): 696.
[13] He K, Zhang X, Ren S, et al. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016: 770.
[14] Wei C, Wang W, Yang W, et al. arXiv: 1808. 04560v1, 2018.
[15] Luo M R, Cui G, Rigg B. Color Research & Application, 2001, 26(5): 340.
[16] Wang K, Wang Z, Zhang L, et al. Current Optics and Photonics, 2020, 4(1): 1.
[17] Wang Z, Bovik A C, Sheikh H R, et al. IEEE Transactions on Image Processing, 2004, 13(4): 600.
[18] Wang S, Zheng J, Hu H-M, et al. IEEE Transactions on Image Processing, 2013, 22(9): 3538.
[19] Li M, Liu J, Yang W, et al. IEEE Transactions on Image Processing, 2018, 27(6): 2828.
