Self-Reference Deep Adaptive Curve Estimation for Low-Light Image Enhancement

Jianyu Wen,Chenhao Wu,Tong Zhang,Yixuan Yu,Piotr Swierczynski

2023-09-10

Abstract:In this paper, we propose a 2-stage low-light image enhancement method called Self-Reference Deep Adaptive Curve Estimation (Self-DACE). In the first stage, we present an intuitive, lightweight, fast, and unsupervised luminance enhancement algorithm. The algorithm is based on a novel low-light enhancement curve that can be used to locally boost image brightness. We also propose a new loss function with a simplified physical model designed to preserve natural images' color, structure, and fidelity. We use a vanilla CNN to map each pixel through deep Adaptive Adjustment Curves (AAC) while preserving the local image structure. Secondly, we introduce the corresponding denoising scheme to remove the latent noise in the darkness. We approximately model the noise in the dark and deploy a Denoising-Net to estimate and remove the noise after the first stage. Exhaustive qualitative and quantitative analysis shows that our method outperforms existing state-of-the-art algorithms on multiple real-world datasets.

Image and Video Processing,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper aims to address the problem of image enhancement under low-light conditions. Specifically, the authors propose a method called Self-Reference Deep Adaptive Curve Estimation (Self-DACE) for low-light image enhancement. The main contributions of this method are as follows:

1. **Novel Unsupervised Loss Function**: Based on the Retinex illumination theory, a new self-referential unsupervised loss function is proposed. This loss function is primarily used for curve-based image enhancement but can also be applied to other neural network methods.

2. **Iterative Low-Light Enhancement Method**: An iterative low-light enhancement method using newly adjusted curves is designed. This method can locally enhance the image without introducing unnecessary artifacts.

3. **Lightweight Convolutional Network**: A lightweight convolutional network suitable for edge devices is proposed, which can be applied in real-time image processing scenarios, such as helping mobile robots perceive their surroundings in low-light environments.

4. **Unsupervised Denoising Method**: An unsupervised denoising method is developed to remove noise that is hidden in the dark and amplified due to brightness enhancement.

5. **Extensive Performance Analysis**: The algorithm undergoes thorough qualitative and quantitative analysis, demonstrating its superior performance on multiple real-world datasets and comparing it with existing state-of-the-art methods.

In summary, the paper proposes a low-light image enhancement method that works effectively even under complex lighting conditions, characterized by its speed and low resource consumption, making it suitable for various application scenarios.