简介：纸为同时的图象分类和注解建议一个新奇概率的生产模型。模型考虑范畴信息能为图象注解提供珍贵信息的事实。一旦一幅图象的范畴被查明，注解词的范围能被缩小，并且产生无关的注解词的概率能被减少。到这个目的，根据班注解图象的想法在模型被介绍。用变化方法，模型的近似推理和参数评价算法被导出，并且为分类并且注解新图象的有效近似也被给。我们的模型的力量在二真实世界数据集上被表明：1600图象LabelMe数据集和1791图象UIUC运动数据集。实验结果证明分类表演与几个最先进的分类模型同样，当注解表演比几个最先进的注解模型的好时。

简介：Forasemi-supervisedclassificationsystem,withtheincreaseofthetrainingsamplesnumber,thesystemneedstobecontinuallyupdated.Asthesizeofsamplessetisincreasing,manyunreliablesampleswillalsobeincreased.Inthispaper,weusefuzzyc-means(FCM)clusteringtotakeoutsomesamplesthatareuseless,andextracttheintersectionbetweentheoriginaltrainingsetandtheclusterafterusingFCMclustering.Theintersectionbetweeneveryclassandclusterisreliablesampleswhichwearelookingfor.Theexperimentresultdemonstratesthatthesuperiorityoftheproposedalgorithmisremarkable.

简介：Thispaperfocusesontheproblemofautomaticimageclassification(AIC)byproposingaframeworkbasedonlatentsemanticanalysis(LSA)andimageregionpairs.Thenovelframeworkemploysrelativespatialarrangementsforregionpairsastheprimaryfeaturetocapturesemantics.Thesignificanceofthispaperistwofold.Firstly,tothebestourknowledge,thisisthefirststudyoftheinfluenceofregionpairsaswellastheirrelativespatialinformationinlatentsemanticanalysisasappliedtoautomaticimageclassification.Secondly,ourproposedmethodforusingtherelativespatialinformationofregionpairsshowgreatpromiseinimprovingimagesemanticclassificationcomparedwiththeclassicallatentsemanticanalysismethodand2Dstringrepresentationalgorithm.