【作者】 龚慧超； 项文波； 侯晓霞； 茅耀斌；

【Author】 GONG Hui-chao XIANG Wen-bo HOU Xiao-Xia MAO Yao-bin(The Collegel of Automation,Nanjing University of Science & Technology,Nanjing 210094)

【机构】 南京理工大学自动化学院；

【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间中的目标概念,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。更多还原

【Abstract】 A irregular segmented region coding algorithm based on Pulse Coupled Neural Network.To screen off inundant erotic information on Internet,a content-based image filtering approach was proposed in this paper.In the approach a skin region segmentation was first performed according to image contents,then effective features were extracted from the segmented region.Ambiguous and abstract concept learning problem in image filtering was turned into a multi-instance learning problem,by which concept target in the feature space of the erotic images could be obtained easily through an effective multi-instance learning algorithm.Furthermore,to acquire the global optimal multi- instance target,an improvement of target searching strategy by simulated annealing algorithm was proposed.The experimental results have demonstrated the effectiveness of the proposed algorithm.更多还原