基于Retinex理论和残差网络的低照度图像增强方法  

Low-light image enhancement method based on Retinex theory and residual network

作　　者：徐轩 姚斌[1] 韩典芝 XU Xuan;YAO Bin;HAN Dianzhi(School of Electronic Information and Artificial Intelligence,Shaanxi University of Science and Technology,Xi′an 710021,China;Shaanxi Joint Laboratory of Artificial Intelligence,Shaanxi University of Science and Technology,Xi′an 710021,China)

机构地区：[1]陕西科技大学电子信息与人工智能学院,西安710021 [2]陕西科技大学陕西省人工智能联合实验室,西安710021

出　　处：《智能计算机与应用》2025年第3期164-169,共6页Intelligent Computer and Applications

基　　金：国家自然科学基金青年项目(61603234)。

摘　　要：针对昏暗环境下拍摄的图像亮度及对比度低的问题,设计了一个基于Retinex理论和深度残差网络的低照度图像增强网络。首先,分解子网络依据Retinex理论分离输入图像的反射分量和照度分量作为后续输入;其次,使用编码器解码器架构的增强网络进行特征提取,通过自适应空间特征融合结构提高特征尺度的不变性;最后,设计了基于残差结构的降噪子网络对反射图去噪,通过残差结构的跳跃连接来弥补传统去噪方法的不足。实验结果表明,本文提出的方法能有效提升图像亮度以及对比度,增强效果细节突出,颜色失真不明显,不仅在主观视觉上有很好的展现,在客观指标上也领先于其他方法。Aiming at the problem of low brightness and contrast of images taken in dark environment,a low illumination image enhancement network based on Retinex theory and depth residual network is designed.Firstly,the decomposition sub-network decomposes the reflection component and illumination component of the input image as the subsequent input according to Retinex theory;Secondly,the enhanced network of encoder and decoder architecture is used for feature extraction,and the invariance of feature scale is improved through adaptive spatial feature fusion structure;Finally,a noise reduction network based on residual structure is designed to denoise the reflection map,and the deficiency of the traditional denoising method is compensated by the jump connection of residual structure.The experimental results show that the method proposed in this paper can improve the brightness and contrast of the image,the enhancement effect details are prominent,the color distortion is not obvious.To sum up,the method not only has a good display in the subjective vision,but also is ahead of other methods in the objective index.

关 键 词：低照度图像增强 RETINEX理论 深度学习 残差网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]