• 首页
  • 期刊简介
  • 编委会
  • 投稿指南
  • 收录情况
  • 杂志订阅
  • 联系我们
引用本文:李天杰,于 欣,宋振英,许立波,万 星.基于高频分量引导生成的古陶器模型纹理修复方法[J].软件工程,2024,(8):70-73.【点击复制】
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  
←前一篇|后一篇→ 过刊浏览
分享到: 微信 更多
基于高频分量引导生成的古陶器模型纹理修复方法
李天杰1, 于 欣2, 宋振英3, 许立波2, 万 星2
(1.浙江理工大学计算机科学与技术学院, 浙江 杭州 310018;
2.浙大宁波理工学院, 浙江 宁波 315100;
3.大连海洋大学, 辽宁 大连 116023)
lllltj@163.com; yuxin@zju.edu.cn; 1553662046@qq.com; xlb@nbt.edu.cn; 13669027591@163.com
摘 要: 古陶器纹理修复是重现历史文物真实面貌的重要手段,对于研究和保护数字文化遗产具有重要意义,同时提供了重要依据。文章针对修复过程中存在的训练数据集规模小、待修复区域大及模型学习困难等问题,提出了一种特殊的修复方法。首先,通过傅里叶滤波器处理输入图像,提取高频分量作为输入条件,为模型提供纹理信息,从而引导修复结果的生成。其次,通过增加颜色损失对模型进行约束,为模型提供颜色信息,使修复结果更加真实。实验证明,与现有图像修复方法相比,该算法在PSNR指标上实现了0.47 dB的提升,同时在SSIM和LPIPS指标上也均表现出不同程度的优化。
关键词: 图像修复;数字文物遗产;高频分量;三维模型
中图分类号: TP391    文献标识码: A
基金项目: 浙江省哲学社会科学规划交叉学科重点支持课题(22JCXK08Z);宁波市自然科学基金项目 (2022J162);宁波市哲学社会科学研究基地项目(JD6-228);宁波市重点技术研发计划研发项目(2022Z167)
Texture Restoration Method for Ancient Pottery Models Based on High-Frequency Component-Guided Generation
LI Tianjie1, YU Xin2, SONG Zhenying3, XU Libo2, WAN Xing2
(1.Zhejiang Sci-Tech University, Hangzhou 310018 China;
2.NingboTech University, Ningbo 315100, China;
3.Dalian Ocean University, Dalian 116023, China)
lllltj@163.com; yuxin@zju.edu.cn; 1553662046@qq.com; xlb@nbt.edu.cn; 13669027591@163.com
Abstract: The restoration of textures on ancient pottery is an important means of reproducing the authentic appearance of historical relics, which provides an important basis and is significant for the research and preservation of digital cultural heritage. Addressing challenges such as the small scale of training datasets, large areas for restoration, and difficulties in model learning during the restoration process, this paper proposes a special restoration method. Firstly, the input image is processed through a Fourier filter, and high-frequency components are extracted as input conditions to provide texture information for the model, thereby guiding the generation of repair results. Secondly, the model is constrained by adding color loss to provide color information for the model to make the repair results more realistic. Experiments show that compared to existing image restoration methods, this algorithm achieves an improvement of 0. 47 dB in the PSNR ( Peak Signal-to-Noise Ratio) indicator, and shows varying degrees of optimization in SSIM (Structural Similarity) and LPIPS (Learned Perceptual Image Patch Similarity) indicators.
Keywords: image restoration; digital cultural heritage; high-frequency component; 3D model


版权所有:软件工程杂志社
地址:辽宁省沈阳市浑南区新秀街2号 邮政编码:110179
电话:0411-84767887 传真:0411-84835089 Email:semagazine@neusoft.edu.cn
备案号:辽ICP备17007376号-1
技术支持:北京勤云科技发展有限公司

用微信扫一扫

用微信扫一扫