一种用于三维模型检索的半监督距离度量学习方法

Method of semi-supervised distance metric learning for 3D model retrieval

在三维模型检索领域,由于语义鸿沟的存在使得无监督的相似匹配技术检索出的结果通常不十分令人满意,而有监督的分类学习方法又常常需要大量的训练样本集。为了在用户提供有限的分类信息下,提高三维模型的检索效率,提出了一种适用于三维模型检索的半监督加权距离度量学习方法。该方法首先通过一种基于图的半监督分类标记繁殖方法增加用户提供的极少量分类信息,随后使用一种改进的加权相关成分分析方法学习一个马氏距离度量,最后将这个学习到的马氏距离度量应用于三维模型检索中。在Princeton Shape Benchm ark上的测试表明,该方法在用户标注模型很少的情况下,检索效果明显好于普通的距离度量方法以及监督的分类学习方法。

In the field of 3D model retrieval,the performance of the unsupervised similitude matching method is not satisfactory because of the semantic gap.However,supervised classification learning method usually needs a lot of training sets.In order to improve the effectiveness of retrieval with a small amount of classification information,a semi-supervised weighted distance metric learning method for 3D model retrieval is introduced.The method uses a graph based semi-supervised label propagation algorithm to increase the very little classification information provided by the user and then adopts an improved weighted relevant component analysis method to learn a Mahalanobis distance function.Then,the Mahalanobis distance metric function can be used to retrieve the 3D models.Experiment results on Princeton Shape Benchmark have shown the effectiveness of our proposed method.