LAE-Net: A locally-adaptive embedding network for low-light image enhancement
作者：
Highlights：
• LAE-Net goes beyond the limitations of previous works and explores the relationship between image characteristics and image quality from the data itself, so as to guide the high-quality image generation.
• The entropy-inspired kernel-selection convolution can adaptively adjust the receptive field size according to its spatial frequency characterized by information entropy.
• The illumination attention transfer sub-net can simultaneously sense global consistency and local details, thereby adjusting the refined features.
• LAE-Net can balance different local enhancement requirements of properties of light intensity, detail presentation and color fidelity, and produce high-quality and visual-pleasing normal light images.
摘要
•LAE-Net goes beyond the limitations of previous works and explores the relationship between image characteristics and image quality from the data itself, so as to guide the high-quality image generation.•The entropy-inspired kernel-selection convolution can adaptively adjust the receptive field size according to its spatial frequency characterized by information entropy.•The illumination attention transfer sub-net can simultaneously sense global consistency and local details, thereby adjusting the refined features.•LAE-Net can balance different local enhancement requirements of properties of light intensity, detail presentation and color fidelity, and produce high-quality and visual-pleasing normal light images.
论文关键词：Locally-adaptive,Image enhancement,Multi-distribution,Image entropy,Kernel selection
论文评审过程：Received 10 January 2022, Revised 15 August 2022, Accepted 6 September 2022, Available online 10 September 2022, Version of Record 15 September 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.109039
