采用两阶段混合策略实现的低照度图像增强算法

doi:10.12263/DZXB.20201018

摘要/Abstract

摘要：

在深入分析现有各主流低照度图像增强(Low Light Image Enhancement, LLIE)算法的基础上，提出了一种采用两阶段混合策略实现的低照度图像增强(Hybrid LLIE, HLLIE)算法.具体地，在第一阶段，对于给定的低照度图像，利用互补效果较好的Fu和Ying两个主流LLIE算法分别对其进行增强预处理，所得到的两张增强后图像称为初步增强图像；在第二阶段，将所得到的两张初步增强图像输入到预先训练好的多通道浅层卷积神经网络(Multi?channel Shallow Convolution Neural Network, MSCNN)模型中，由MSCNN模型将两张初步增强图像优化组合为一张具有更高图像质量的最终增强图像.实验结果表明：与各主流LLIE算法相比，所提出的HLLIE算法在各个客观图像质量评价指标上有显著优势，人工主观评价亦能证实这一点.

关键词: 低照度图像增强, 两阶段混合策略, 初步增强图像, 多通道浅层卷积神经网络, 组合

Abstract:

After carefully analyzing the characteristics of existing low light image enhancement (LLIE) algorithms, we proposed a hybrid LLIE (HLLIE) algorithm with two-stage hybrid strategy. Specifically, in the first stage, for a given low-light image, we chose the complementary LLIE algorithms, i.e., Fu and Ying, to enhance the low-light image, respectively. In the second stage, the two preliminary enhanced images were used as the inputs of the multi-channel shallow convolution neural network (MSCNN), and then the pre-trained MSCNN conducted optimal combination of the two preliminary enhanced images to achieve further enhancement regarding the image quality. The experimental results show that, compared with the state-of-the-art LLIE algorithms, the HLLIE algorithm has obvious advantages in terms of the objective image quality metrics on the low-light images, and the subjective evaluation also can confirm this.

Key words: low-light image enhancement, two-stage hybrid strategy, initial enhanced image, multi-channel shallow convolution neural network, combination

中图分类号:

TP391.41

徐少平, 陈孝国, 李芬, 林珍玉, 陈晓军. 采用两阶段混合策略实现的低照度图像增强算法[J]. 电子学报, 2021, 49(11): 2166-2170.

Shao-ping XU, Xiao-guo CHEN, Fen LI, Zhen-yu LIN, Xiao-jun CHEN. A Low‑Light Image Enhancement Algorithm Using Two‑Stage Hybrid Strategy[J]. Acta Electronica Sinica, 2021, 49(11): 2166-2170.

图/表 7

参考文献 18

1	Wang Y F, Liu H M, Fu Z W. Low-light image enhancement via the absorption light scattering model[J]. IEEE Transactions on Image Processing, 2019, 28(11):5679 - 5690.
2	Ren Y R, Ying Z Q, Li T H, et al. LECARM: Low-light image enhancement using camera response model[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2019, 29(4):968 - 981.
3	Guo Y H, Ke X, Ma J, et al. A pipeline neural network for low-light image enhancement[J]. IEEE Access, 2019, 7: 13737 - 13744.
4	Park S, Yu S, Kim M, et al. Dual autoencoder network for Retinex-based low-light image enhancement[J]. IEEE Access, 2018, 6:22084 - 22093.
5	Cheng H D, Shi X J. A simple and effective histogram equalization approach to image enhancement[J]. Digital Signal Processing, 2004, 14(2):158 - 170.
6	Fu X Y, Zeng D L, Yue H, et al. A weighted variational model for simultaneous reflectance and illumination estimation[A]. 2016 IEEE Conference on Computer Vision and Pattern Recognition[C]. USA:IEEE, 2016. 2782 - 2790.
7	Ying Z Q, Ge L, Ren Y R, et al. A new image contrast enhancement algorithm using exposure fusion framework[A]. International Conference on Computer Analysis of Images and Patterns[C]. Sweden:Springer Verlag, 2017.36 - 46.
8	Loh Y P, Liang X F, Chan C S. Low-light image enhancement using Gaussian process for features retrieval[J]. Signal Processing: Image Communication, 2019, 74: 175 - 190.
9	Guo X J, Yu L, Ling H B. LIME: Low-light image enhancement via illumination map estimation[J]. IEEE Transactions on Image Processing, 2017, 26(2):982 - 993.
10	徐少平,刘婷云,林珍玉,等.深度卷积神经网络降噪模型的技术瓶颈与研究展望[J].中国图象图形学报, 2019, 24(8):1207 - 1214.
	Xu S P, Liu T Y, Lin Z Y, et al. Main bottlenecks and research prospects of the deep convolutional neural network-based denoising model[J].Journal of Image and Graphics, 2019, 24(8):1207 - 1214.(in Chinese)
11	Zhang K, Zuo W M, Chen Y J, et al. Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising[J]. IEEE Transactions on Image Processing, 2017, 26(7):3142 - 3155.
12	Cai J R, Gu S H, Zhang L. Learning a deep single image contrast enhancer from multi-exposure images[J]. IEEE Transactions on Image Processing, 2018, 27(4): 2049 - 2062.
13	Fu X Y, Liao Y H, Zeng D L, et al. A probabilistic method for image enhancement with simultaneous illumination and reflectance estimation[J]. IEEE Transactions on Image Processing, 2015, 24(12):4965 - 4977.
14	Li M D, Liu J Y, Yang W H, et al. Structure-revealing low-light image enhancement via robust Retinex model[J]. IEEE Transactions on Image Processing, 2018, 27(6):2828 - 2841.
15	Fu X Y, Zeng D L, Yue H, et al. A fusion-based enhancing method for weakly illuminated images[J]. Signal Processing, 2016, 129: 82 - 96.
16	Wang Z, Li Q. Information content weighting for perceptual image quality assessment[J]. IEEE Transactions on Image Processing, 2011, 20(5):1185 - 1198.
17	Ying Z Q, Ge L, Wen G. A bio-inspired multi-exposure fusion framework for low-light image enhancement[J]. Journal of Latex Class Files, 2015, 14(8):1 - 10.
18	Kai Z, Ma K D, Hassen R, et al. Perceptual evaluation of multi-exposure image fusion algorithms[A]. International Workshop on Quality of Multimedia Experience[C]. Singapore: IEEE, 2014. 7 - 12.

组合模式		组合模式
Case1	Fu+LIME	Case2	Fu2016+LIME
Case3	Fu+Fu2016	Case4	Fu+Li
Case5	Li+LIME	Case6	Fu2016+Li
Case7	Li+Ying	Case8	Li+HVS
Case9	Fu+HVS	Case10	Fu+Ying
Case11	Fu2016+HVS	Case12	LIME+HVS
Case13	Fu2016+Ying	Case14	HVS+Ying
Case15	LIME+Ying	-	-

组合模式		组合模式
Case1	Fu+LIME	Case2	Fu2016+LIME
Case3	Fu+Fu2016	Case4	Fu+Li
Case5	Li+LIME	Case6	Fu2016+Li
Case7	Li+Ying	Case8	Li+HVS
Case9	Fu+HVS	Case10	Fu+Ying
Case11	Fu2016+HVS	Case12	LIME+HVS
Case13	Fu2016+Ying	Case14	HVS+Ying
Case15	LIME+Ying	-	-

评价指标	组合模式
评价指标	Case1	Case2	Case3	Case4	Case5	Case6	Case7	Case8
IW‑PSNR	23.6894	22.0059	21.3225	22.5234	22.3017	21.5752	22.5093	21.6897
IW‑SSIM	0.9417	0.9156	0.9123	0.9294	0.9215	0.9181	0.9354	0.9135
	Case9	Case10	Case11	Case12	Case13	Case14	Case15	-
IW‑PSNR	22.6713	24.3390	23.0180	22.2188	21.9148	23.4783	21.7534	-
IW‑SSIM	0.9298	0.9510	0.9368	0.9183	0.9219	0.9441	0.9203	-

评价指标	组合模式
评价指标	Case1	Case2	Case3	Case4	Case5	Case6	Case7	Case8
IW‑PSNR	23.6894	22.0059	21.3225	22.5234	22.3017	21.5752	22.5093	21.6897
IW‑SSIM	0.9417	0.9156	0.9123	0.9294	0.9215	0.9181	0.9354	0.9135
	Case9	Case10	Case11	Case12	Case13	Case14	Case15	-
IW‑PSNR	22.6713	24.3390	23.0180	22.2188	21.9148	23.4783	21.7534	-
IW‑SSIM	0.9298	0.9510	0.9368	0.9183	0.9219	0.9441	0.9203	-

评价指标	Conv+BN+ReLU网络层重复的次数
评价指标	1	3	5	7	9	12	15
IW-PSNR	22.89	21.28	21.50	21.48	22.27	22.43	23.07