基于注意力机制的特征融合的暗图像增强网络  

Dark Image Enhancement Network Based on Feature Fusion with Attention Mechanism

作　　者：宋涛[1] 黄微[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200444

出　　处：《工业控制计算机》2023年第4期112-114,共3页Industrial Control Computer

摘　　要：暗图像能见度低、颜色失真、细节丢失,会对计算机视觉应用造成干扰。提出了基于无参注意力、高效通道注意力、像素注意力和通道注意力的特征融合的暗图像增强网络。提出的网络使用对抗式训练,具有提高图像亮度、增强图像颜色内容和细节信息的能力。算法增强结果的结构相似度(PSNR)和峰值信噪比(SSIM)均优于所对比的算法。Dark images have low visibility,color distortion and loss of details,which can interfere with computer vision applications.This paper proposes a dark image enhancement network based on feature fusion of parameterless attention,efficient channel attention,pixel attention,and channel attention.The proposed network uses adversarial training and has the ability to improve image brightness and enhance image color content and detail information.The structural similarity(PSNR)and peak signal-to-noise ratio(SSIM)of the enhanced results of the proposed algorithm are better than those of the compared algorithms.

关 键 词：暗图像增强 无参注意力 高效通道注意力 像素注意力 通道注意力 特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]