期刊文献+

基于双残差卷积网络的低照度图像增强 被引量:4

Low-light image enhancement based on dual-residual convolutional network
下载PDF
导出
摘要 为解决当前低照度图像增强问题,提出了一种基于双残差卷积网络的图像增强算法。首先,根据Retinex理论模型,将正常照度图像合成低照度图像,再分别将它们分解在R(红)、G(绿)、B(蓝)3个分量上,然后通过特征提取模块和双残差模块学习低照度图像与正常照度图像在各分量的映射关系,获得各分量上的增强图像,最后合成增强的RGB图像。采用双边滤波优化增强的RGB图像,使得所获得的图像更加接近参考图像。实验表明,本文所提算法,对于处理合成的低照度图像,峰值信噪比最高可达25.9311 dB,结构相似度最高可达0.9452;对于处理真实的低照度图像,盲图像质量评估指标高于其他算法,且运行速度更快。 In order to solve the current problem about low-light image enhancement,an algorithm of image enhancement based on dual-residual convolutional network is proposed.First,according to Retinex theory,the normal-light image is synthesized into low-light image,the synthetic is decomposed onto the three components of R、G and B,and learning the mapping relations between low-light image and normal-light image on all components through the module of feature extraction as well as dual-residual.Then the enhanced image on all components can be obtained,and finally the enhanced RGB image is synthesized.Subsequently,the bilateral filtering is used to optimize the enhanced RGB image so that the obtained image is analogical to the reference image.The experiment results show that the algorithm proposed in this paper,whose Peak Signal to Noise Ratio can reach up to 25.9311 dB and Structural Similarity Index can reach up to 0.9452 in terms of processing synthesized low-light image,and whose novel blind image quality assessment can exceed other compared algorithms and the algorithm in this paper goes faster in terms of processing real low-light image.Therefore,the proposed algorithm is superior to the contrast algorithms.
作者 陈清江 屈梅 CHEN Qing-jiang;QU Mei(School of Science,Xi’an University of Architecture and Technology, Xi’an 710055, China)
出处 《液晶与显示》 CAS CSCD 北大核心 2021年第2期305-316,共12页 Chinese Journal of Liquid Crystals and Displays
基金 国家自然科学基金(No.61403298) 陕西省自然科学基金(No.2015JM1024)。
关键词 低照度图像增强 双残差网络 特征提取 RETINEX理论 low-light image enhancement dual-residual network feature extraction Retinex theoretical model
  • 引文网络
  • 相关文献

参考文献7

二级参考文献32

共引文献136

同被引文献30

引证文献4

二级引证文献15

相关主题

;
使用帮助 返回顶部