基于视觉显著的图像拷贝检测关键技术研究-信息安全专业论文.docx

著区域的显著图。该算法不仅取得了较好的显著区域检测结果，而且可以提供全分辨率或高分辨率的显著图同时有效易于实现。此外，在显著区域检测算法基础上，构建了一个非显著阈值依赖同时可以提取具有清晰轮廓的显著物体的检测框架。在该框架下，提出了一种基于显著区域差异的显著密度最大算法用于搜索包含显著物体的矩形窗口。该算法降低了搜索到的包含显著物体的矩形窗中属于背景的像素，改善了利用GrabCut进行显著物体提取的结果。（3）提出了一种椭圆显著区域Log协方差描述子ESRCOV（EllipticallySalientRegionlog-COVariance）。该描述子通过显著图融合、显著区域椭圆拟合、协方差矩阵构建和Log变换构建而成。和现有的基于显著性的区域协方差描述子相比，ESRCOV较为全面的考虑了图像中的显著性，并在更利于特征描述的椭圆上进行区域协方差的构建。ESRCOV在取得较好的综合拷贝检测性能的同时特征维数更加紧致。关键词：视觉感知；显著性检测；显著区域检测；全局特征；椭圆显著区域协方差；AbstractDuetotherapiddevelopmentofmediaprocessing,computernetworkandmobilecomputingtechnologies,visualmediaconsistingofimagesandvideoshaveshownanexplosivegrowth.Atthesametime,themethodsofvisualmediaaccess,edit,publish-ingandtransferhavealsoundergoneagreatchange.Copyingmediacontentandcopiestransferalsohasbecomemoreeasilyandwidely.Thedevelopmentofcloudcomputinghasincreasedcopiestransfer,whichhasleadhugecopiedstoredinthecloud.Therightsofmediacopyrightholdersareinfringed.Ontheotherhand,storageresourcesarewast-ed.Therefore,e?ectivelydetectingmediacopieshasbecameahotresearchissueandattractedmuchattentionfromindustrialandacademicresearchers.Content-basedcopydetectioniswidelystudiedtodetectmediacopies.Thegoalofcontent-basedcopydetectionistomakethecomputershavetheabilityofidentifyingmediacontentashumanbeings.Thehumanvisualperceptionsystemfortheunderstand-ingofmediacontentandtheabilitytoidentifymediaisfarbeyondthecurrentallmediaprocessingtechnologies.Therefore,thisdissertationisbasedonsaliency-basevisualat-tentionresearchtobuildanovelglobalfeatureforimagecopydetection.Inordertomeasureoverallimagesaliency,thisdissertationcarriesonresearchtodetecteye-fixationbasedsaliencyandsalientregions.Saliencydetectionresearchcansolvethekeyissueofextractingimageperceptualcontent.Anotherkeyissueofsaliency-basedimagecopydetectionishowtobuildarobustglobalfeature.Basedonsaliencyresearchinthisdis-sertation,anovelsaliency-basedglobalfeaturewithgoodrobustperformanceisbuiltbytheresearchonregioncovariancematrixdescriptor.Themainworkandcontributionsofthisdiss