简介：ＭｏｔｉｏｎＥｓｔｉｍａｔｉｏｎＢａｓｅｄｏｎＧｌｏｂａｌａｎｄＬｏｃａＣｏｍｐｅｎｓａｂｉｌｉｔｙＡｎａｌｙｓｉｓTXＭｏｔｉｏｎＥｓｔｉｍａｔｉｏｎＢａｓｅｄｏｎＧｌｏｂａｌａｎｄＬｏｃａｌＣｏｍｐｅｎｓａｂｉｌｉｔｙＡｎａｌｙｓｉｓＦａｎＪｉｎ...

简介：Thispaperpresentsanewmethodfordetectionofedgesindigitalangiographicimages.Itisfoundthatvariancesoflocalregionsacrossedgesofimagesarestatisticallydifferentfromthatofthosewherenoedgeiscrossed.Thisdifferencecanbeutilizedforthedetectionofedgesofangiographicimages.Analgorithmbasedonlocalvarianceisproposed.Asaresult,theedge-detectionalgorithmisnotsensitivetonoiseandlow-leveltexturesofimages.Acomputerprogrambasedonthenewalgorithmhasbeendevelopedandusedbyseveralhospitals.

简介：编码基于的方法的一篇小说说出同样本地的二进制取向代码(LBOCode)因为palmprint识别被建议。palmprint图象第一与Gabor过滤器的一个银行被卷，然后取向信息与一条winner-take-all规则被达到。随后，产生取向印射数组被一致本地二进制模式操作。因此，LBOCode图象被完成它在象素水平包含palmprint取向信息。进一步，我们把LBOCode图象划分成几相等尺寸并且nonoverlapping区域，和摘录从每个区域的统计代码直方图独立地，它在地区性的水平造对palmprint的全球描述。在匹配舞台，在二palmprints之间的匹配的分数被精明的二张空间提高的直方图的不同完成，它带计算简洁的利益。试验性的结果证明建议方法与几个最先进的方法的相比完成更有希望的识别性能。

简介：因为质地图象不能被单个象素的灰色的水平信息直接处理，我们建议反映在每个象素集中的补丁的紧张分发的一个新质地描述符。然后，Potts模型的一般多相的图象分割模型被增加质地描述符的区域信息为质地分割扩大。一个快数字计划基于裂口Bregman方法被设计加快计算过程。算法是有效的，并且质地描述符和典型函数能容易被实现。用合成质地图象，真实自然景色图象和合成的孔雷达图象的实验被介绍给在我们的方法和另外的最先进的技术之间的质的比较。结果证明我们的方法能精确地分割目标区域并且在分割自然图象特别与另外的方法相比是竞争的。

简介：Inthispaper,alocal-learningalgorithmformulti-agentispresentedbasedonthefactthatindividualagentperformslocalperceptionandlocalinteractionundergroupenvironment.Asforin-dividual-learning,agentadoptsgreedystrategytomaximizeitsrewardwheninteractingwithenvi-ronment.Ingroup-learning,localinteractiontakesplacebetweeneachtwoagents.Alocal-learningalgorithmtochooseandmodifyagents'actionsisproposedtoimprovethetraditionalQ-learningalgorithm,respectivelyinthesituationsofzero-sumgamesandgeneral-sumgameswithuniqueequi-libriumormulti-equilibrium.Andthislocal-learningalgorithmisprovedtobeconvergentandthecomputationcomplexityislowerthantheNash-Q.Additionally,throughgrid-gametest,itisindicatedthatbyusingthislocal-learningalgorithm,thelocalbehaviorsofagentscanspreadtoglobe.

简介：Inthispaper,anovelframework,namedasglobal-localfeatureattentionnetworkwithrerankingstrategy(GLAN-RS),ispresentedforimagecaptioningtask.Ratherthanonlyadoptingunitaryvisualinformationintheclassicalmodels,GLAN-RSexplorestheattentionmechanismtocapturelocalconvolutionalsalientimagemaps.Furthermore,weadoptrerankingstrategytoadjustthepriorityofthecandidatecaptionsandselectthebestone.TheproposedmodelisverifiedusingtheMicrosoftCommonObjectsinContext(MSCOCO)benchmarkdatasetacrosssevenstandardevaluationmetrics.ExperimentalresultsshowthatGLAN-RSsignificantlyoutperformsthestate-of-the-artapproaches,suchasmultimodalrecurrentneuralnetwork(MRNN)andGoogleNIC,whichgetsanimprovementof20%intermsofBLEU4scoreand13pointsintermsofCIDERscore.

简介：Anovelactivecontourmodelisproposed,whichincorporateslocalinformationdistributionsinafuzzyenergyfunctiontoeffectivelydealwiththeintensityinhomogeneity.Moreover,theproposedmodelisconvexwithrespecttothevariablewhichisusedforextractingthecontour.Thismakesthemodelindependentontheinitialconditionandsuitableforanautomaticsegmentation.Furthermore,theenergyfunctionisminimizedinacomputationallyefficientwaybycalculatingthefuzzyenergyalterationsdirectly.Experimentsarecarriedouttoprovetheperformanceoftheproposedmodeloversomeexistingmethods.Theobtainedresultsconfirmtheefficiencyofthemethod.