基于决策树C4.5集成算法的图像自动标注  

Image annotation based on decision tree C4.5 ensemble algorithm

作　　者：张华忠[1] 侯进[2] Zhang Huazhong;Hou Jin(Aviation Engineering Institute,Civil Aviation Flight University of China,Guanghan Sichuan 618307,China;Dept.of Information Science ＆ Technology,Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]中国民用航空飞行学院航空工程学院,四川广汉618307 [2]西南交通大学信息科学与技术学院,成都611756

出　　处：《计算机应用研究》2018年第7期2222-2224,共3页Application Research of Computers

基　　金：国家自然科学基金面上项目(61371165);中国民用航空飞行学院高等学校面上项目(XM0327;XM2272);中国民用航空飞行学院高等学校青年基金资助项目(Q2012-051)

摘　　要：针对决策树C4.5集成算法中基分类器多样性差的问题,提出了修正矩阵correction matrix-C4.5(CMC4.5)集成学习算法,并将其应用于图像自动标注。该算法首先对特征子集进行多样性处理,然后通过构造修正矩阵依次得到基分类器C4.5全新训练数据集,实现训练数据集之间的多样性和属性特征完整性,完成集成算法。对比实验表明,CMC4.5集成学习方法较大地提高了分类准确率。将CMC4.5集成学习与图像标注相结合,实现了基于CMC4.5的图像自动标注。Aiming at the problem of diversity in the base classifier of C4. 5 ensemble algorithm,this paper presented a correction matrix-C4. 5（ CMC4. 5） ensemble algorithm and applied in image automatic annotation. Firstly,it randomly divided the feature set into subsets and then handled the subsets. Secondly,it obtained the new training set of the base classifier C4. 5 by correction matrix. Finally,it completed the ensemble algorithm,which had the diversity of training set and attribute characteristics of integrity. The experiments show that CMC4. 5 makes a positive and effective contribution to classification performance. It realized image automatic annotation based on CMC4. 5 by combining CMC4. 5 with image annotation.

关 键 词：C4.5算法 集成学习 修正矩阵 图像标注 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]