摘要
针对Retinex-Net算法存在放大噪声、颜色失真、光晕伪影的问题,提出一种基于Retinex理论和多尺度注意力的低光图像增强算法。首先引入一个光照平滑度损失函数,能更好地学习分解反射分量和光照分量。其次,构建一个基于多尺度注意力模块的反射分量恢复网络,以处理退化的反射分量。然后,采用轻量型的U-Net结构进一步增强光照分量。最后,将恢复的反射分量和增强的光照分量融合重构,得到最终的增强图像。实验结果表明,该算法相比于LIME、MBLLEN、Retinex-Net、KinD等算法性能更优。
Due to the problems of noise amplification,color distortion and halo artifacts in Retinex Net algorithm,a low light image enhancement algorithm based on Retinex theory and multi-scale attention is proposed.The algorithm firstly introduces an illu-mination smoothness loss function,which can better learn to decompose the reflectance component and the illumination component.Secondly,a reflectance recovery network based on multi-scale attention module is constructed,which aims to deal with the degradation of the decomposed reflection components.Then a lightweight U-Net structure is used to enhance the illumination components.Finally,the recovered reflection component and the enhanced illumination component are fused and reconstructed to obtain the final enhanced image.Experimental results show that the proposed algorithm has a better performance than LIME、MBLLEN、Ret-inex-Net,KinD,etc.
作者
郭业才
张丹
王雪
GUO Yecai;ZHANG Dan;WANG Xue(School of Electronics&Information Engineering,Nanjing University of Information Science&Technology,Nanjing 210044)
出处
《计算机与数字工程》
2025年第8期2314-2318,2335,共6页
Computer & Digital Engineering
