AMDCENR: Attention Mechanism and Deep Curve Estimation under Noise Reduction Network for low-light image enhancement

Photographs captured under low-light conditions are often plagued with issues such as low brightness, poor contrast and noise, making them unsuitable for various image processing applications. To address these challenges, previous low-illumination image enhancement methods, based on the zero-reference depth network model (Zero-DCE), have attempted to approximate pixel-wise and higher-order curves by iteratively applying themselves. However, these methods have not considered the inevitable presence of noise in low-light images, which can have a significant impact on the quality of the enhanced images. To address this limitation, we present an approach, named Attention Mechanism and Deep Curve Estimation under Noise Reduction Network (AMDCENR-net), for low-light image enhancement. Our approach utilizes a regularization model and elemental subtraction to iteratively refine the light map and achieve image denoising. To improve the quality of low-light images and address the issue of missing information, we augment our approach by incorporating channel attention and spatial attention. Channel attention aims to enhance the informative channels in the image while suppressing the unimportant ones. Spatial attention, on the other hand, focuses on highlighting the salient regions of the image. Extensive experiments demonstrate the advantages of our method over state-of-the-art methods qualitatively and quantitatively.