Low contrast and weak detail features of images collected in a low-light environment will seriously affect the accuracy and stability of machine vision detection. In recent years, the low-light image enhancement technology has made remarkable progress. However, the existing low-light image enhancement algorithms have some problems, such as image detail loss, low brightness, local exposure, insufficient visual naturalness, complex algorithm and high resource overhead. To solve the above problems, a low-light image enhancement algorithm based on extended atmospheric scattering model is proposed. Firstly, the maximum value of R, G and B color channels is calculated and the initial transmission map is obtained by gamma correction. Secondly, the main structure and fine structure of the initial transmission map were extracted, PCA (Principal Component Analysis) method was used to fuse the main structure transmission map and fine structure transmission map to obtain the optimized local consistency transmission map of texture detail removal. Then, the inverse atmospheric light value is calculated using the dark pixel of the bright channel. Finally, the LIEAS (Low-light Image Enhancement via Extend Atmospheric Scattering Model) model is solved to obtain the final enhanced image with natural color and good contrast. The enhanced model derived by the algorithm is similar to the Retinex enhanced model in form, but the difference is that there is an additional correction term in the LIEAS model, which can better suppress the excessive enhancement and detail loss in the image. The algorithm uses the image fusion method to optimize the transmission image and can reproduce the contour and texture details well. In order to evaluate the algorithm objectively, spatial frequency, average gradient, edge intensity and natural image quality evaluation are used as the image quality evaluation metrics. In order to verify the effectiveness of the algorithm, the parameter analysis experiment, model analysis experiment and performance comparison experiment are carried out respectively. In the parameter analysis experiment, firstly, the influence of the selection of gamma parameters on the enhanced image is analyzed. The subjective visual analysis and objective data analysis are carried out on the test results under different parameters, and a good gamma parameter value is obtained. Secondly, the influence of the selection of the maximum filtering window size of the bright channel on the solution of the inverse atmospheric light value and the enhanced image is analyzed. The test results under different window sizes are analyzed to obtain an appropriate window size. Then, the darkest pixel proportion of the bright channel in the solution of the inverse atmospheric light value is selected and analyzed. Finally, this paper verifies the advantages of the transmission map optimization method based on fusion technology. In the model analysis experiment, compared with Retinex model, spatial frequency and average gradient of the proposed algorithm are significantly improved, which also has prominent visual advantages, indicating that the correction term in the proposed algorithm can better suppress the excessive enhancement and detail loss of the enhanced image, and has good enhancement ability. At the same time, by changing the atmospheric light value in the model, the proposed algorithm can also be used in image dehazing, and the image dehazing can get a good effect from both subjective and objective aspects. In the performance verification experiment, three low-light image datasets were selected to test, and the performance of the proposed enhancement algorithm was compared with that of other eight algorithms from both subjective and objective aspects. Compared with the other eight algorithms, this algorithm has the advantages of bright background, high contrast, complete edge details, natural, vivid image, avoiding local overexposure and so on. The algorithm has more advantages in spatial frequency, average gradient, and edge intensity, which indicates that the algorithm has better performance in the aspects of image color richness and image sharpness. Both for the metric analysis of the whole image of the dataset and for the metric analysis of a single image, the proposed algorithm is very advantageous, and the model is simple and low complexity. Compared with the existing enhancement algorithms, the proposed algorithm has some advantages in detail information retention, contrast enhancement, image naturalness and local overexposure suppression.
为验证本文所提出的基于融合技术的传输图优化方法的性能,对比分析了几种常见的传输图优化方法,包括导向滤波器(Guided Filter,GF)、双边滤波器(Bilateral Filter,BF)和加权最小二乘法(Weighted Least Square,WLS),不同优化方法对应的增强结果如图8所示。
图 8. 不同优化方法下的增强图像和传输图
Fig. 8. Enhanced image and transmission map under different optimization methods
高动态范围图像(High Dynamic Range Imaging,HDR)是在计算机图形学与电影摄影术中用来实现比普通数位图像技术更大曝光动态范围(即更大的明暗差别)的一组技术,其目的是正确地表示真实世界中从太阳光直射到最暗阴影这样大的范围亮度。类似的,WEI K等[25]介绍了如何构造更真实的噪声数据,能够合成更好的匹配图像以及形成具有物理特性的真实样本,他提出了一种方法来校准现有现代数码相机的噪声参数。受此启发,为验证本文模型的性能,对高动态范围图像进行测试,并对比了一种HDR图像增强模型方法(Tone-map First Denoise Last,TFDL)[26],增强结果如图10所示。
图 10. TFDL与LIEAS模型的增强图像比较
Fig. 10. Comparison of enhanced image between TFDL and LIEAS model
[2] GUO Xiaojie, LI Yu, LING Haibin. LIME: low-light image enhancement via illumination map estimation[J]. IEEE Transactions on Image Processing, 2016, 26(2): 982-993.
[3] REN Yurui, YING Zhengqiang, LI T H, et al. LECARM: low-light image enhancement using the camera response model[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2018, 29(4): 968-981.
[4] REN Xutong, YANG Wenhan, CHENG W H, et al. LR3M: robust low-light enhancement via low-rank regularized retinex model[J]. IEEE Transactions on Image Processing, 2020, 29: 5862-5876.
[5] WANG Manli, TIAN Zijian, GUI Weifeng, et al. Low-light image enhancement based on nonsubsampled shearlet transform[J]. IEEE Access, 2020, 8: 63162-63174.
[6] MUSTAFAW A, YAZIDH, KHAIRUNIZAMW, et al. Image enhancement based on discrete cosine transforms (DCT) and discrete wavelet transform (DWT): a review[C]. IOP Conference Series: Materials Science and Engineering, 2019, 557(1): 012027.
[8] JIANG Yifan, GONG Xinyu, LIU Ding, et al. Enlightengan: deep light enhancement without paired supervision[J]. IEEE Transactions on Image Processing, 2021, 30: 2340-2349.
[9] LI Jingjiang, FENG Xiaoming, HUA Zhen. Low-light image enhancement via progressive-recursive network[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 31(11): 4227-4240.
[10] LEE H, SOHN K, MIN D. Unsupervised low-light image enhancement using bright channel prior[J]. IEEE Signal Processing Letters, 2020, 27: 251-255.
[11] DONGXuan, PANGYi, WENJiangtao. Fast efficient algorithm for enhancement of low lighting video[M]. ACM Siggraph 2010 Posters, 2010: 1-1.
[12] FENG Xiaoming, LI Jingjiang, HUA Zhen. Low-light image enhancement algorithm based on an atmospheric physical model[J]. Multimedia Tools and Applications, 2020, 79(43): 32973-32997.
[13] WANG Yunfei, LIU Heming, FU Zhaowang. Low-light image enhancement via the absorption light scattering model[J]. IEEE Transactions on Image Processing, 2019, 28(11): 5679-5690.
[14] HE Kaiming, SUN Jian, TANG Xiaoou. Single image haze removal using dark channel prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 33(12): 2341-2353.
[15] XU Li, YAN Qiong, XIA Yang, et al. Structure extraction from texture via relative total variation[J]. ACM Transactions on Graphics, 2012, 31(6): 1-10.
[16] LEEH, JEONJ, KIMJ, et al. Structure‐texture decomposition of images with interval gradient[C]. Computer Graphics Forum, 2017, 36(6): 262-274.
[17] WANG Shuhang, ZHENG Jin, HU H M, et al. Naturalness preserved enhancement algorithm for non-uniform illumination images[J]. IEEE Transactions on Image Processing, 2013, 22(9): 3538-3548.
[18] FUXueyang, ZENGDelu, HUANGYue, et al. A weighted variational model for simultaneous reflectance and illumination estimation[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016: 2782-2790.
[20] FU Xueyang, ZENG Delu, HUANG Yue, et al. A fusion-based enhancing method for weakly illuminated images[J]. Signal Processing, 2016, 129: 82-96.
[21] LU Kun, ZHANG Lihong. TBEFN: a two-branch exposure-fusion network for low-light image enhancement[J]. IEEE Transactions on Multimedia, 2020, 23: 4093-4105.
[22] WANG Liwen, LIU Zhisong, SIU W C, et al. Lightening network for low-light image enhancement[J]. IEEE Transactions on Image Processing, 2020, 29: 7984-7996.
[23] GUOChunle, LIChongyi, GUOJichang, et al. Zero-reference deep curve estimation for low-light image enhancement[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 1780-1789.
[24] MITTAL A, SOUNDARARAJAN R, BOVIK A C. Making a “completely blind” image quality analyzer[J]. IEEE Signal Processing Letters, 2012, 20(3): 209-212.
[25] WEIKaixuan, FUYing, YANGJiaolong, et al. A physics-based noise formation model for extreme low-light raw denoising[C]. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 2758-2767.
[26] HULitao, CHENHuaijin, ALLEBACHJ P. Joint multi-scale tone mapping and denoising for HDR image enhancement[C]. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2022: 729-738.
[27] LEE C, LEE C, KIM C S. Contrast enhancement based on layered difference representation of 2D histograms[J]. IEEE Transactions on Image Processing, 2013, 22(12): 5372-5384.