Underexposed Photo Enhancement Using Deep Illumination Estimation

By Ruixing Wang, Qing Zhang, Chi-Wing Fu, Xiaoyong Shen, Weishi Zheng, Jiaya Jia. Computer Vision and Pattern Recognition, 2019

957 citations recorded by the source corpus.

This paper presents a new neural network for enhancing underexposed photos. Instead of directly learning an image-to-image mapping as previous work, we introduce intermediate illumination in our network to associate the input with expected enhancement result, which augments the network's capability to learn complex photographic adjustment from expert-retouched input/output image pairs. Based on this model, we formulate a loss function that adopts constraints and priors on the illumination, prepare a new dataset of 3,000 underexposed image pairs, and train the network to effectively learn a rich variety of adjustment for diverse lighting conditions. By these means, our network is able to recover clear details, distinct contrast, and natural color in the enhancement results. We perform extensive experiments on the benchmark MIT-Adobe FiveK dataset and our new dataset, and show that our network is effective to deal with previously challenging images.

Subfields: Atomic and Molecular Physics, and Optics, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/8062f3536fec7647548af53adc191069458fdc9c

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Atomic and Molecular Physics, and Optics

In Computer Vision and Pattern Recognition

Citations in the corpus

Benchmarking Low-Light Image Enhancement and Beyond (2021, background): “In someof these works (Lore et al. 2017; Cai et al. 2018; Wang et al. 2019), efforts are put into creating paired low/normal-light datasets for network training.”
Low-Light Image Enhancement With Semi-Decoupled Decomposition (2020, methodology): “In recent years, Convolutional Neural Network (CNN) has been broadly applied [20]–[22], where skip connection is used to preserve”
Attention Guided Low-Light Image Enhancement with a Large Scale Low-Light Simulation Dataset (2019, methodology): “[77] [73] [8]”
Learning Image-Adaptive 3D Lookup Tables for High Performance Photo Enhancement in Real-Time (2020, methodology): “Method FiveK (480p) FiveK (original) PSNR SSIM 4E∗ PSNR SSIM 4E∗ UPE [10] 21.”
Learning a Simple Low-Light Image Enhancer from Paired Low-Light Instances (2023, background): “To correct contrast, uncover textures, and remove sensor noise, great efforts have been made in developing Low-light Image Enhancement (LIE) algorithms in the past decades [1, 5, 6, 8, 28, 35].”
Learning to Restore Low-Light Images via Decomposition-and-Enhancement (2020, methodology): “(c) WVM [14] (d) DeepUPE [40] (e) CAPE [25] (f) LIME [20]”
Learning Enriched Features for Fast Image Restoration and Enhancement (2022, background): “Similar to [119], [120], [121], we also consider the enhanced images of expert C as the ground-truth.”
Learning to Enhance Low-Light Image via Zero-Reference Deep Curve Estimation (2021, methodology): “[6], RetinexNet [7], LightenNet [39], MBLLEN [23]), and one GANbased method (EnlightenGAN [9]).”
Retinex-inspired Unrolling with Cooperative Prior Architecture Search for Low-light Image Enhancement (2020, methodology): “[34] designed an imageto-illumination network architecture based on the bilateral learning framework.”
Zero-Reference Deep Curve Estimation for Low-Light Image Enhancement (2020, methodology): “CNN-based models usually use self-captured paired data for network training [27, 5, 24] while GAN-based models elaborately select unpaired data [10].”
From Fidelity to Perceptual Quality: A Semi-Supervised Approach for Low-Light Image Enhancement (2020, methodology): “These methods [20, 4, 2, 24, 27] learn to remove the related composite degradation based on the well-prepared paired low/normal-light images and elaborately designed models.”
SNR-Aware Low-light Image Enhancement (2022, methodology): “Following the setting in [40], we evaluate results via user ratings on the six questions shown in Fig.”
Guided Curriculum Model Adaptation and Uncertainty-Aware Evaluation for Semantic Nighttime Image Segmentation (2019, background): “learns jointly from stylized images and unlabeled real images of increasing darkness and exploits the prior knowledge from a map. There is a rich literature on low-light image enhancement [27], [28], [29], which is also relevant to our work. However, its focus is more on the low-level goal of visual”
Sparse Gradient Regularized Deep Retinex Network for Robust Low-Light Image Enhancement (2021, methodology): “In [15], local and global information is jointly captured to adjust the illumination of the input low-light images.”
Low-Light Image Enhancement with Normalizing Flow (2021, methodology): “2017; Cai, Gu, and Zhang 2018), l1 loss(Cai, Gu, and Zhang 2018), structural similarity (SSIM) (Cai, Gu, and Zhang 2018), smoothness loss (Wang et al. 2019a; Zhang et al. 2019) and color loss (Wang et al.”
Toward Fast, Flexible, and Robust Low-Light Image Enhancement (2022, neutral): “Compared with adopting a direct mapping between the low-light observation and illumination (a commonly-used pattern in existing works, e.g., [14,22]), learning a residual representation substantially reduces the computational difﬁculty to 1 The architecture for H θ will be explored in Sec.”
Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement (2023, methodology): “Similar to [15, 18, 49], we regard R as a well-exposed image.”
Beyond Brightening Low-light Images (2021, methodology): “For the deep learning methods, including Retinex-Net (Wei et al. 2018), GLAD (Wang et al. 2018), KinD (Zhang et al. 2019) and KinD++, we have retrained their models based on the 4500 pair images from the FiveK (Pic# 0001-4500).”
Learning Enriched Features for Real Image Restoration and Enhancement (2020, methodology): “69 dB improvement over DeepUPE[100] on the Adobe-Fivek dataset.”