基于扩散模型的轮廓生成式信息隐藏方法

打开文本图片集

中图分类号：TP309.7 文献标志码：A 文章编号：2095-2945(2025)30-0134-04

Abstract:Aimingattheproblemofpoorqualityofsecretimagesingenerativeinformationhiding methods，this paper proposesacontourgenerativeinformationhiding methodbasedondifusionmodel.Thesenderfrstextractscontourpointsfrom theimage，buildsacontourcandidate listbasedonthecontourpoints，andthenconstructsaHuffmancode，selects correspondingcontourpointsaccordingtothesecretinformationusesadiferentiablerastertogenerateaseriesofontourlines fromallcontourpoints,andrendersthecontourlinestogenerateasecretcontourimage.Inordertoensurethatthedense contourimagecanwellretainthecontourfeaturesoftheoriginalimage,CLIPmodelisusedtooptimizeit.Then,thedensity carryingcontourimageisgeneratedbycombiningadifusionmodelwithoutclasifcationguidance.Afterthereceiverobtainsthe secretimage,itfirstextractsthesecretimage,thenextractsthecontourpointsfromitandconstructsaHuffmancodetoextract thesecretinformation.Experimentalresultsshowthattheoverallperformanceofthismethodisbetterthanthecurrent mainstreamgenerativeimagehidingschemes，especiallythegeneratedhigher-qualitycryptography-carryingimages，andthe extraction accuracy and robustness can be taken into account.

Keywords: generative information hiding;difusion model; CLIP model; secret image; hiding method

随着互联网的快速发展，信息技术在经济、文化、军事和医疗等诸多领域发挥着重要作用。（剩余7613字）