摘要
针对传统的高层语义提取方法对训练数据集的高度依赖以及准确率不足的问题,提出一种基于Adaboost-SVM的高层次语义概念提取方法,将SVM作为Adaboost的弱分类器训练方法,并充分利用Adaboost对训练数据的平衡及融合弱分类器的特点,提取出高可靠的语义检测器。实验结果表明,与传统方法相比,该方法不仅跨越训练数据不平衡的障碍,而且能够提取出更加可靠的语义检测器。
To tackle the problems of traditional high-level semantic extraction methods that they are highly dependent upon training data sets and they suffer from inaccuracy,an Adaboost-SVM based high-level semantic concept extraction method is proposed that takes SVM as Adaboost＇s weak classifier training method,fully takes advantage of Adaboost＇s characteristics like training data balancing and integration of weak classifiers,and extracts a highly reliable semantic detector.Experimental results show that,compared with traditional methods,the novel approach not only steps over the barrier of unbalanced training data,but also attains a more reliable semantic detector.
出处
《计算机应用与软件》
CSCD
北大核心
2012年第4期24-26,56,共4页
Computer Applications and Software
基金
国家自然科学基金项目(61001148)
作者简介
高荣星，硕士生，主研领域：多媒体视频信息处理与网络完全。
