Abstract: In order to decrease the “semantic gap” between low-level visual features and high-level semantics, a novel method of low-level visual features mapping high-level semantics is proposed, which uses a tree structure Support Vector Machine（SVM）. Binary decision tree is utilized to design SVM classifiers and distance measure is used to realize clustering at the kernel space. Binary decision tree architecture can shorten the time of semantic classification so much, while separating the semantic classes which have bigger distance can promise high classification accuracy. The experimental results show that proposed method not only can promise high accuracy, but also shorten retrieval time so much.

Key words: image semantic classification, binary decision tree, Support Vector Machine（SVM）

摘要： 为了减小低层视觉特征和高层语义之间存在的“语义鸿沟”，提出一种采用树结构支持向量机实现图像底层视觉特征到高层语义的映射方法。利用二叉树结构构建支持向量机（SVM），在SVM核函数空间利用距离作为树节点处的分类度量。二叉树的结构可以大大减小语义分类的时间，而将距离较大的语义类先分离开保证了语义分类具有较高的准确率。实验证明，该方法在保证准确率的同时可以在较大程度上缩短分类检索时间。

关键词: 图像语义分类, 二叉树, 支持向量机