基于条件互信息量的随机蕨特征匹配算法

摘要

为解决随机蕨属性组合独立性假设导致分类器性能降低的问题,提出了一种基于条件互信息量的随机蕨特征识别方法,通过已知类别属性组合之间的互信息量最大化,将关联度大的特征属性划分为一个蕨丛,并建立朴素贝叶斯模型,训练分类器.此特征属性选择方法改进了随机蕨离线训练机制,有效提高了分类器的性能,显著改善了随机蕨特征匹配算法的有效性.结合实例及仿真分析,验证了所提算法在保证实时性的同时提高了特征匹配的精确度.%To overcome the deficiencies of the assumption of independence, which restricted random Ferns's classifying performance greatly, a algorithm on keypoint recognition using random Ferns based on conditional mutual information is presented. By picking features attributes which maximize their mutual information with the class to predict conditional to any feature already picked, it ensures joining attributes with close relationship into a ferns. Naive Bayesian model is designed to train the classifier. The performance of random Ferns classifier is improved effectively. The experimental results prove that this method improves the precisions while the amount of computation do not have notable change.