基于图元变换的建筑彩绘纹样图像矢量化方法

打开文本图片集

关键词：图像矢量化；图像模式；残差网络；图像分类；可微图像合成中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）07-038-2206-07doi：10.19734/j.issn.1001-3695.2024.09.0374

Abstract：Tovectorizearchitecturalpaintedpaternimagesthatcontainreusablegraphicelements，whilepreservingtheindependenceandtransformationparametersof thereusable graphic elements，thispaperproposedavectorizationmethodforarchitecturalpaintedpaternimagesbasedonelementtransformation.Thismethodfirstlydividedcomplexpattersintolocalpatterns，thenconductedmulti-classelementdetectiononthelocalpaternstocompleteelementfilteringandinitializeelement transformationparameters.Subsequently，itusedanimproveddiferentiablesythesisalgorithmtocalculatethetransforation parametersof thelements，ultimatelyachievingvectorizationofthepaternimagewhileretainingthetransformationparameters of thereusablegraphicelements.Experimentalresultsdemonstratethatthis methodcanachieveasmalerimagereconstruction eroronarchitecturalpaintedpatterndatasets，whilealsopreservingthetransformationparametersofthereusablegraphic elements.Basedonthecharacteristicsof thepainted paters，thispaperproposedavarietyofvectorizationmetrics tocompare the performance of various methods.Thismethodhas advantages in terms ofreconstruction accuracyandtheretentionof element transformation parameters，and can be applied to the vectorization of images with reusable graphic elements.

Key words：image vectorization；image patterns；residual network；image classfication；diferential compositing

0 引言

图像矢量化是将位图转换为由数学参数描述的矢量图的一项技术，矢量图相较于位图具有分辨率无关性与占用空间小的优点。（剩余17064字）