Object Categorization in Context based on Probabilistic Learning of Classification Tree with Boosted Features and Co-occurrence Structure
Object Categorization in Context based on Probabilistic Learning of Classification Tree with Boosted Features and Co-occurrence Structure
复制标题
基于增强特征和共现结构的分类树概率学习的上下文中的对象分类
DOI：
10.1007/978-3-642-41914-0_41
复制
发表时间：
2013
期刊：
影响因子：
--
通讯作者：
Masayasu Atsumi
中科院分区：
文献类型：
--
作者：
毛利 貴之;杉町 勇和;東藤 大樹 岩崎 敦;横尾 真;Masayasu Atsumi
提出了一种基于上下文的概率对象分类方法，通过学习具有提升特征和共现结构的分类树来实现。在该方法中，对每个场景类别获取对象类，为所有对象类生成具有提升特征的分类树，并分析场景中对象类之间的共现性。在识别中，基于在共现约束下使用复合提升特征的分类树搜索来同时确定场景中的对象类别，并且基于场景类别的对象类别组成来推断前景对象。通过使用图像数据集中的多个类别的图像的实验，它示出了对象分类性能的提高，通过使用提升的特征和共生结构，特别是通过使用它们两者。
This paper proposes a probabilistic method of object categorization in context through learning a classification tree with boosted features and co-occurrence structure. In this method, object classes are obtained for each scene category, a classification tree with boosted features is generated for all the object classes and co-occurrence is analyzed among object categories in scenes. In recognition, object categories in a scene are simultaneously determined based on a classification tree search using composite boosted features under co-occurrence constraint and a foreground object is inferred based on object category composition of scene categories. Through experiments using images of plural categories in an image data set, it is shown that object categorization performance is improved by using boosted features and co-occurrence structure, especially by using both of them.