WMANet: Wavelet-Based Multi-Scale Attention Network for Low-Light Image Enhancement

This paper put forward a deep-learning model for LLIE. Our key idea is to construct a frequency pyramid of low-light images in terms of a wavelet transform, starting from the global color information of the image and gradually enhancing the image details. To this end, for different frequency components, we constructed a low-frequency restoration network (LRN) and a high-frequency recon-struction network (HRN) to enhance the low-frequency and high-frequency information, respectively. Moreover, we introduce a simplified attention mechanism (SA) and a feature extraction block (FEB) in different networks and demonstrate their effectiveness in ablation experiments. Additionally, we constructed a new dataset consisting of 4500 multi-scale expo-sure image pairs to compensate for the shortcomings of the existing datasets. This enables the proposed model to recover precise details, sharp contrasts, and vivid colors in complex low-light scenes. Extensive experiments prove that our proposed method produces more convincing results than the existing LLIE approaches. In the future work, we will address the problem of unprocessed sensor noise and challenging imaging under extremely dark conditions.

[1]   N. Sharma, V. Kumar, and S. K. Singla, ‘‘Single image defogging using deep learning techniques: Past, present and future,’’ Arch. Comput. Methods Eng., vol. 28, no. 7, pp. 4449–4469, Dec. 2021. [2]   Y.-L. Liu, W.-S. Lai, Y.-S. Chen, Y.-L. Kao, M.-H. Yang, Y.-Y. Chuang, and J.-B. Huang, ‘‘Single-image HDR reconstruction by learning to reverse the camera pipeline,’’ in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Seattle, WA, USA, Jun. 2020, pp. 1648–1657. [3]   K. Singh, R. Kapoor, and S. K. Sinha, ‘‘Enhancement of low exposure images via recursive histogram equalization algorithms,’’ Optik, vol. 126, no. 20, pp. 2619–2625, Oct. 2015. [4]   P. Gupta, J. Kumare, U. Singh, and R. Singh, ‘‘Histogram based image enhancement techniques: A survey,’’ Int. J. Comput. Sci. Eng., vol. 5, no. 6, pp. 475–484, 2017. [5]   P. Babakhani and P. Zarei, ‘‘Automatic gamma correction based on average of brightness,’’ Adv. Comput. Sci., Int. J., vol. 4, pp. 156–159, Nov. 2015. [6]   S. Rahman, M. M. Rahman, M. Abdullah-Al-Wadud, G. D. Al-Quaderi, and M. Shoyaib, ‘‘An adaptive gamma correction for image enhancement,’’ EURASIP J. Image Video Process., vol. 2016, no. 1, pp. 1–13, Dec. 2016. [7]   E. H. Land, ‘‘An alternative technique for the computation of the designator in the retinex theory of color vision,’’ Proc. Nat. Acad. Sci. USA, vol. 83, no. 10, pp. 3078–3080, 1986. [8]   L. Wang, G. Fu, Z. Jiang, G. Ju, and A. Men, ‘‘Low-light image enhancement with attention and multi-level feature fusion,’’ in Proc. IEEE Int. Conf. Multimedia Expo Workshops (ICMEW), Shanghai, China, Jul. 2019, pp. 276–281. [9]   J. W. Cooley and J. W. Tukey, ‘‘An algorithm for the machine calculation of complex Fourier series,’’ Math. Comput., vol. 19, no. 90, p. 297, Apr. 1965. [10]   J. Huang, Y. Liu, F. Zhao, K. Yan, J. Zhang, Y. Huang, M. Zhou, and Z. Xiong, ‘‘Deep Fourier-based exposure correction network with spatial-frequency interaction,’’ in Proc. Eur. Conf. Comput. Vis., 2022, pp. 163–180.