基于AFSVM-MIL算法的图像标注  

Image annotation based on AFSVM-MIL arithmetic

作　　者：邓剑勋[1] 熊忠阳[1] 曾代敏[2] 

机构地区：[1]重庆大学计算机学院,重庆400044 [2]重庆大学数理学院,重庆400044

出　　处：《计算机应用研究》2011年第10期3917-3919,3924,共4页Application Research of Computers

基　　金：中国博士后科学基金资助项目(20070420711);中央高校基本科研业务费科研专项--研究生科技创新基金资助项目(CDJXS11180001);中央高校基本科研业务费科研专项--自然科学类资助项目(CDJZR10100023);重庆市科委自然科学基金计划资助项目(2007BB2372)

摘　　要：通常情况下关键字只标注在图像上,而多示例(MIL)检索的需要将关键字下沉到区域。针对这个问题,在模糊支持向量机算法(FSVM)的基础上提出了一种改进的自适应模糊支持向量机多示例学习算法(AFS-VM-MIL算法),在多示例学习的框架下把区域级的图像标注变成了一种有监督的学习。该方法利用AFSVM-MIL对训练集进行分类,结合包之间的相似度进行广义集合运算,可以有效地将关键字进行下沉,从而达到减少人工标注工作量的目的。实验结果表明,该方法有效且性能优于其他方法。In most existing training data set,keywords are usually associated with images instead of regions.And not only supervised learning methods but also multi instance learning（MIL） methods need to associate Key words with regions.So this paper brought forward an improved MIL arithmetic（AFSVM-MIL）,which based on active learning fuzzy support vector machine.Then image annotation problem became a supervised learning one in MIL framework.This method used AFSVM-MIL to classify the training datasets.And do generalized set operation on datasets by the similarity between individual image and class center point.So the Key words were labeled on the instance（the sub area of image）.Thus the aims of project were achieved.Experimental results on the ECCV2008 image set show that this algorithm is feasible,and the performance is superior to other MIL algorithms.

关 键 词：图像标注 多示例学习 自适应模糊支持向量机 广义集合运算 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]