电子学报 ›› 2021, Vol. 49 ›› Issue (1): 72-76.DOI: 10.12263/DZXB.20191286

• 学术论文 • 上一篇    下一篇

采用深度学习与图像融合混合实现策略的低照度图像增强算法

徐少平, 林珍玉, 张贵珍, 陈孝国, 李芬   

  1. 南昌大学信息工程学院, 江西南昌 330031
  • 收稿日期:2019-11-18 修回日期:2020-05-11 出版日期:2021-01-25 发布日期:2021-01-25
  • 通讯作者: 徐少平
  • 作者简介:林珍玉 女,1996年2月出生于江西省九江市.现为南昌大学硕士研究生,主要研究方向为图像处理与计算机视觉.E-mail:401030918076@email.ncu.edu.cn
  • 基金资助:
    国家自然科学基金(No.61662044,No.61163023,No.51765042);江西省自然科学基金(No.20171BAB202017)

A Low-Light Image Enhancement Algorithm Using the Hybrid Strategy of Deep Learning and Image Fusion

XU Shao-ping, LIN Zhen-yu, ZHANG Gui-zhen, CHEN Xiao-guo, LI Fen   

  1. School of Information Engineering, Nanchang University, Nanchang, Jiangxi 330031, China
  • Received:2019-11-18 Revised:2020-05-11 Online:2021-01-25 Published:2021-01-25
  • Supported by:
     

摘要: 提出了一种采用深度学习与图像融合混合实现策略的低照度图像增强算法.首先,利用照射分量预测模型直接基于输入的低照度图像快速地估计出其最佳照射分量并在Retinex模型框架下获得一张整体上适度曝光图像;其次,将低照度图像本身及它的过曝光图像作为适度曝光图像的修正补充图像参与融合;最后,采用局部结构化融合和色度加权融合机制技术将制备好的3张待融合图像进行融合以获得最终的增强图像.实验数据表明:本文算法相较于各种主流对比算法在主客观图像质量评价指标上均有显著优势,在局部图像结构细节上具有更好的边缘保持和颜色保真效果.

 

关键词: 低照度图像增强, 深度学习, 照射分量预测模型, 适度增强图像, 过曝光图像, 图像融合

Abstract: An improved low-light image enhancement (LLIE) algorithm based on the hybrid strategy of deep learning and image fusion was proposed in this paper.We first adopted illumination prediction model to quickly estimate the optimal illumination component from a given low-light image and obtain its corresponding moderately exposed image within the framework of the Retinex model.Then the low-light image and its over-exposed image were used as supplementary images for the moderately exposed image.Finally,the three images were fused within the framework of the local structured fusion and the chrominance weighted fusion mechanism to obtain the final enhanced image.Experimental results demonstrate that,compared with the state-of-the-art LLIE algorithms,the proposed hybrid strategy has significant advantages in both subjective and objective image quality evaluation metrics with better image edge preservation and color fidelity effect on local image details.

Key words: low-light image enhancement, deep learning, illumination predict model, moderately exposed image, over-exposed image, image fusion

中图分类号: