一种面向手绘草图的多阶段跨域图像检索方法

打开文本图片集

关键词：手绘草图检索；跨域图像转换；多特征提取；相似度融合

中图分类号：TP391 文献标志码：A 文章编号：1001-3695（2025）12-039-3832-09

doi：10.19734/j. issn. 1001-3695.2024.12.0538

Multi-stage cross-domain image retrieval method based on hand-drawn sketches

Wang Zhixian’，Lei Haopeng1，Wang Mingwen’，Luo Guoliang²，Fu Yirui1 （1.Schoolofe&otnilUsit；el teractive Technology，East China Jiaotong University，Nanchang 33oo13，China）

Abstract：Toaddress thechallengesinsketch-basedimageretrievaltasks，wherethelackoftextureandcolordetailsinsketchesmakesross-modalfeaturealigmentdiicult，andwhereiformationlossuringross-domain transformationaffectsetrieval accuracy，thispaperproposedamulti-stagecros-domainimageretrieval methodbasedonhand-drawnsketches （MSCD）.The method was divided into three stages.Firstly，the methodusedagenerative model to transform hand-drawn sketchesintonaturalimages，supplementingvisualdetailfeaturesandreducingthestylisticdiferencesbetweensketchesand naturalimages.Secondly，themethoddesignedamulti-branchfeatureextractionnetworktocomprehensivelycaptrestructural andsemanticinformationfromsketches，generatedimages，andrealnaturalimages.Finally，itusedasimilarity-weightedfusionstrategytointegratethecomplementaryinformationfromsketchesandgeneratedimages，enhancingtheaccuracyanddiversityofthefialretrievalresults.Extensiveexperimentsontwopublicdatasets，SketchyExtandTU-BerlinExt，demostrate that MSCDachievessignificantimprovements inmAPandPrec@1OOoverexistingadvanced methods，validating theeffectivenessofMSCD in sketch retrieval tasks.

Keywords：hand-drawnsketchesretrieval；cross-domainimagetransformation；multi-featureextraction；similarityfusior

0 引言

随着平板电脑、智能手机等触控设备的广泛普及，用户能够更加便捷地通过电子设备绘制手绘草图，将构思迅速转换为视觉表达。（剩余25023字）