视频语义分类特征选择算法  

Algorithm of Feature Selection for Semantic Video Classifier

作　　者：魏维[1] 赵学龙[1] 刘凤玉[1] 许满武[2] 

机构地区：[1]南京理工大学计算机科学与技术系,南京210094 [2]南京大学计算机科学与技术系,南京210008

出　　处：《系统仿真学报》2006年第5期1143-1146,共4页Journal of System Simulation

基　　金：国家自然科学基金项目资助(60273035);江苏省科技攻关项目(BE2003064)

摘　　要：提出一种将连续特征数字量化后进行特征选择的算法(ABFSA)。利用样本集中的先验信息选择出特征值域中最具类间区分意义的区域,将其作为完整量化区间。采用向后式的启发搜索策略,搜索合并后能使贝叶斯分类错误率降低的相邻量化区间。合并搜索得到的两相邻量化区间,量化的级数降低一阶。重复搜索和合并过程,直至贝叶斯分类错误率不再降低为止。所有特征搜索、合并完成后,总的特征量化阶数得到大幅降低。用UCI仿真数据集及真实视频数据进行实验,对比结果表明该算法能有效选取视频语义概念分类的重要特征,其综合性能较优。Classifier performance could be improved by selecting the most important features from high dimension feature set. A method of classification based on approximated Bayesian error feature selection algorithm （ABFSA） was proposed.In order to get the distinctive value partition of feature, the neighborhood of mean for each class was preserved. Then, the residual value field of feature was divided. Backward sequential feature selection, a heuristic search strategy was used in the merging step of quantity partition. At the same time, quantity complexity was remarkably reduced. The results of experiments comparing to other feature selection methods indicate the algorithm can effectively select features.

关 键 词：特征选择 特征量化 贝叶斯分类错误率 视频语义概念 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]