《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (4): 1240-1247.DOI: 10.11772/j.issn.1001-9081.2022030479

• 多媒体计算与计算机仿真 • 上一篇    

结合注意力的双分支残差低光照图像增强

祖佳贞(), 周永霞, 陈乐   

  1. 中国计量大学 信息工程学院,杭州 310018
  • 收稿日期:2022-04-13 修回日期:2022-06-07 接受日期:2022-06-14 发布日期:2023-04-11 出版日期:2023-04-10
  • 通讯作者: 祖佳贞
  • 作者简介:周永霞(1975—),男,浙江诸暨人,副教授,博士,主要研究方向:机器视觉、人工智能;
    陈乐(1998—),男,浙江杭州人,硕士研究生,主要研究方向:图像处理。
  • 基金资助:
    浙江省自然科学基金资助项目(LY19F030013)

Dual-branch residual low-light image enhancement combined with attention

Jiazhen ZU(), Yongxia ZHOU, Le CHEN   

  1. College of Information Engineering,China Jiliang University,Hangzhou Zhejiang 310018,China
  • Received:2022-04-13 Revised:2022-06-07 Accepted:2022-06-14 Online:2023-04-11 Published:2023-04-10
  • Contact: Jiazhen ZU
  • About author:ZHOU Yongxia, born in 1975, Ph. D., associate professor. His research interests include machine vision, artificial intelligence.
    CHEN Le, born in 1998, M. S. candidate. His research interests include image processing.
  • Supported by:
    Natural Science Foundation of Zhejiang Province(LY19F030013)

摘要:

在低光条件下拍摄的照片会因曝光不足而产生一系列的视觉问题,如亮度低、信息丢失、噪声和颜色失真等。为了解决上述问题,提出一个结合注意力的双分支残差低光照图像增强网络。首先,采用改进InceptionV2提取浅层特征;其次,使用残差特征提取块(RFB)和稠密残差特征提取块(DRFB)提取深层特征;然后,融合浅层和深层特征,并将融合结果输入亮度调整块(BAM)调整亮度,最终得到增强图像。同时,结合注意力机制设计特征融合块(FFM)捕获重要的特征信息,以帮助恢复低光照图像的暗部区域。此外,引入一个联合损失函数从多方面衡量网络训练损失。实验结果表明,相较于鲁棒的视网膜大脑皮层模型(RRM)、Zero-DCE(Zero-Reference Deep Curve Estimation)和EnlightenGAN(Enlighten Generative Adversarial Network),在LOL(LOw-Light)数据集上,所提网络的峰值信噪比(PSNR)指标分别提高了49.9%、40.0%和18.5%;在LOL-V2数据集上,结构相似性(SSIM)指标分别提高了20.3%、50.0%和34.5%。所提网络在提高低光照图像亮度的同时降低了噪声,减少了颜色失真和伪影,得到的增强图像更加清晰自然。

关键词: 低光照, 图像增强, 注意力机制, 双分支, 联合损失函数

Abstract:

Photos taken in low-light conditions will suffer from a series of visual problems due to underexposure, such as low brightness, loss of information, noise and color distortion. In order to solve the above problems, a dual-branch residual low-light image enhancement network combined with attention was proposed. Firstly, the improved InceptionV2 was used to extract shallow features. Secondly, Residual Feature extraction Block (RFB) and Dense RFB (DRFB) were used to extract deep features. Thirdly, the shallow and deep features were fused, and the fusion result was input into Brightness Adjustment Module (BAM) to adjust the brightness, and finally the enhanced image was obtained. At the same time, a Feature Fusion Module (FFM) was designed in combination with attention mechanism to capture important feature information, which helped to restore dark areas in low-light images. In addition, a joint loss function was introduced to measure the network training loss from multiple aspects. Experimental results show that, compared with RRM (Robust Retinex Model), Zero-DCE (Zero-reference Deep Curve Estimation) and EnlightenGAN (Enlighten Generative Adversarial Network), on LOL (LOw-Light) dataset, the Peak Signal-to-Noise Ratio (PSNR) indicator of the proposed network is increased by 49.9%, 40.0% and 18.5% respectively. Meanwhile, the Structural Similarity Index Measure (SSIM) indicator of the proposed network is increased by 20.3%, 50.0% and 34.5% compared with those of the above three on LOL?V2 dataset. The proposed network improves the brightness of low-light images while reducing noise, color distortion and artifacts, resulting in sharper and more natural enhanced images.

Key words: low-light, image enhancement, attention mechanism, dual-branch, joint loss function

中图分类号: