基于多示例学习的不良内容图像过滤算法的研究.pdfVIP

龚慧超等：基于多示例学习的不良内容图像过滤算法研究 基于多示例学习的不良内容图像过滤算法研究 龚慧超项文波侯晓霞茅耀斌 (南京理工大学自动化学院，南京，210094，ghccow@163．corn) 摘要：针对目前互联网中存在许多不良信息，本文讨论了基于图像内容的信息过滤技术。论文对图像做基于内容的肤 色分割．在肤色区域内提取图像的有效特征，将图像过滤中模糊、抽象的概念学习问题转化为多示例学习问题，在多 示例学习框架下来求得敏感图像在特征空问中的目标概念，进而利用此目标概念对图像进行过滤。此外，论文还采用 模拟退火算法对多示例中目标概念的求解问题提出了改进，并通过真实的数据对算法有效性进行了验证。 关键词：图像过滤多示例特征提取模拟退火算法 onMulti-Instancebasedectionable Study Learningobj ImageFiltering GONG Yao—bin Hui-chao，XIANGWen-bo，HOUXiao-Xia，MA0 (The UniversityofScience＆Technology,Na可mg210094) College，ofAutomation，Nanjing onPulse Neara】Network．Toscreenoff Abstraet：A based irregularsegmentedregioncodingalgorithm Coupled inundanteroticinformationon content-based was i11this the Intemet，a imagefilteringapproachproposedpaper．In askin wasfirst to contents．thelleriectivefeatureswere approachregionsegmentationperformedaccordingimage in was extractedfromthe andabstract turned segmentedAmbiguous conceptlearningproblemimagefiltering region intoamulti．instance which inthefeature oftheerotic couldbe learningproblem。byconcepttarget space images obtained aneffectivemulti—instance the multi- easilythrough learningalgorithm．Furthermore。toacquireglobaloptimal of simulated was instance improvementsearchingstrategyby target．an target annealingalgorithmproposed．111e resul