摘要
视频语义分类中常遇到多峰正态分布属性,如采用单峰值正态分布设计的贝叶斯分类模型会造成较大分类误差。本文采用定步长组合划分算法(FLCPA)对多峰分布属性值域按类进行划分,以留一校验法(LOOCV)估算分类错误,找出给定步长下属性的多峰分布边界点,并用监督参数估计推断出每个分段区间上的概率分布函数,从而得到整个值域上的总体分布。此外,文中给出了涉及多峰分布属性的视频语义分类器设计步骤。实验数据表明,该方法能明显降低分类错误,有效提高分类性能。
Attributes with multi-normal distribution are common in classifier design for video-semantic concept. In this case, a model assuming that the value of attributes for each class is normally distributed with some mean will lead to poor classification performance. In the paper, an approach based on fixed-length combination partition algorithm （FLCPA） is presented in the partition of attribute value-field. Leave-one-out cross-validation （LO（R2V） is used to estimate classifier error. In addition, the detail of classifier design about multi-normality distribution attribute is given. The result of experiment indicate the method could reduce classifier error and improve classifier performance.
出处
《计算机科学》
CSCD
北大核心
2006年第4期111-114,共4页
Computer Science
基金
国家自然科学基金(60273035)
江苏省科技攻关项目(BE2003064)资助
