基于典型相关分析和距离度量学习的零样本学习

发布时间：2018-03-17 01:26

本文选题：零样本学习　切入点：典型相关分析　出处：《天津大学学报(自然科学与工程技术版)》2017年08期 　论文类型：期刊论文

【摘要】：零样本学习是一类特殊的图像分类问题,是指测试数据的类别在训练数据中没有出现的情况.为了更好地描述语义特征空间中图像特征和语义特征的距离关系,本文将距离度量学习引入零样本学习任务.具体而言,首先利用典型相关分析将样本的图像特征和相应类别的语义特征映射至公共特征空间;然后,利用距离度量学习衡量图像特征和语义特征之间的距离;最后,使用最近邻分类器进行分类.通过在流行的Aw A和CUB数据集中的实验,证明了所提方法的有效性和鲁棒性.
[Abstract]:Zero sample learning is a special image classification problem, which refers to the situation in which the category of test data does not appear in the training data, in order to better describe the distance relationship between image features and semantic features in semantic feature space. In this paper, distance metric learning is introduced into the zero-sample learning task. Firstly, the image features of the samples and the semantic features of the corresponding categories are mapped to the common feature space by canonical correlation analysis. Distance metric learning is used to measure the distance between image features and semantic features. Finally, the nearest neighbor classifier is used to classify images. The effectiveness and robustness of the proposed method are proved by experiments in popular Aw A and CUB datasets.
【作者单位】：天津大学电气自动化与信息工程学院;
【基金】：国家自然科学基金资助项目(61472273,61632018)~~
【分类号】：TP391.41

【相似文献】