融合混合空洞卷积和动态卷积的敦煌壁画修复  

Dunhuang mural inpainting combining hybrid dilated convolution and dynamic convolution

作　　者：刘仲民[1,3] 李耀龙 胡文瑾[2] LIU Zhong-min;LI Yao-long;HU Wen-jin(College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China;College of Mathematic and Computer Science,Northwest Minzu University,Lanzhou 730030,China;Key Laboratory of Gansu Advanced Control for Industrial Processes,Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]兰州理工大学电气工程与信息工程学院,甘肃兰州730050 [2]西北民族大学数学与计算机科学学院,甘肃兰州730030 [3]兰州理工大学甘肃省工业过程先进控制重点实验室,甘肃兰州730050

出　　处：《计算机工程与设计》2025年第2期595-602,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(62061042);兰州理工大学甘肃省工业过程先进控制重点实验室开放基金项目(2022KX10)。

摘　　要：为有效修复壁画破损区域的语义信息、解决壁画深层特征信息提取不足导致的修复伪影以及修复失真等问题,提出一种融合混合空洞卷积与动态卷积的敦煌壁画修复模型。针对修复伪影问题,在模型编码阶段设计一种混合残差模块;针对修复失真问题,通过在动态核预测分支和动态语义及图像滤波分支中加入动态卷积来提高网络的预测和滤波性能。实验结果表明,所提模型具有更高的评价指标,且视觉效果上具有更细致的纹理,语义信息更丰富,边缘结构更连贯。To effectively inpaint the semantic information of damaged areas of murals and to solve problems such as inpainting artifacts and inpainting distortions caused by insufficient extraction of deep feature information of murals,a Dunhuang mural inpainting model that combined hybrid dilated convolution and dynamic convolution was proposed.To solve the problem of inpainting artifacts,a hybrid residual module was designed in the encoding stage of the model.To solve the problem of inpainting distortion,dynamic convolution was added to the dynamic kernel prediction branch and the dynamic semantic and image filtering branches to improve the prediction and filtering performance of the network.Experimental results show that the proposed model has higher evaluation indicators,more detailed visual texture,richer semantic information,and more coherent edge structure.

关 键 词：信息处理技术 壁画修复 混合空洞卷积 动态卷积 图像滤波 残差网络 深度学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]