基于深度学习算法的低照度寺院壁画图像增强研究  

Low-Light Enhancement in Temple Mural Images Based on Deep Learning Algorithm

作　　者：田欢[1] Tian Huan(Lanzhou Vocational and Technical College,Lanzhou 730070,China)

机构地区：[1]兰州职业技术学院,甘肃兰州730070

出　　处：《黑河学院学报》2024年第5期136-139,共4页Journal of Heihe University

基　　金：2022年兰州市科技计划项目“基于计算机视觉的兰州市文物资源三维重建方法与数字化应用研究”(2022-2-77);2023年甘肃省自然科学基金项目“甘肃非遗刺绣数字化保护技术研究”(23JRRA1471)。

摘　　要：随着图像增强技术被越来越多地应用于文物数字化保护,诸多低照度图像增强算法的改进研究越来越受到重视。针对采用机器学习等方法导致图像分解结果不适定的问题,构建Retinex-Pro网络模型,改进对低照度壁画图像进行增强的方法。在有效保证获得图像照度增强效果的同时,使用较少网络参数,结合注意力机制有效提升处理速度,避免产生过拟合使结果表现更为平滑,且细节部分更加清晰。经实验证实,该方法可有效应用于大量寺院壁画类文物的低照度图像增强任务。With the wide application of image enhancement technology in the digital protection of cultural relics,more and more attention has been paid to the improvement of low-light image enhancement algorithms.To solve the problem of ill-posed image decomposition results caused by machine learning and other methods,a Retinex-Pro network model is constructed to improve the method of enhancing low illumination mural images.For eff ectively ensuring the image illumination enhancement eff ect,the measures are suggested to take:using fewer network parameters,improving the processing speed with the help of attention mechanism,avoiding overfi tting to make the result smoother and the details clearer.Experimental results show that this method can be eff ectively applied to the low-light image enhancement of a large number of temple murals.

关 键 词：低照度图像 RETINEX 深度神经网络 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]