A Pipeline Neural Network for Low-Light Image Enhancement

• Main Authors: Yanhui Guo, Xue Ke, Jie Ma, Jun Zhang
• Format: Article
• Language: English
• Published: IEEE 2019-01-01
• Series: IEEE Access
• Subjects: Convolutional neural network; low-light image enhancement; LLIE-Net
• Online Access: https://ieeexplore.ieee.org/document/8607964/

Low-light image enhancement is an important challenge in computer vision. Most of the low-light images taken in low-light conditions usually look noisy and dark, which makes it more difficult for subsequent computer vision tasks. In this paper, inspired by multi-scale retinex, we present a low-light image enhancement pipeline network based on an end-to-end fully convolutional networks and discrete wavelet transformation (DWT). First, we show that multiscale retinex (MSR) can be considered as a convolutional neural network with Gaussian convolution kernel, and blending the result of DWT can improve the image produced by MSR. Second, we propose our pipeline neural network, consisting of denoising net and low-light image enhancement net, which learns a function from a pair of dark and bright images. Finally, we evaluate our method both in the synthetic dataset and public dataset. The experiments reveal that in comparison with other state-of-the-art methods, our methods achieve a better performance in the perspective of qualitative and quantitative analyses.