基于共享特征相对属性的零样本图像分类

发布时间：2018-02-24 15:08

本文关键词： 相对属性多任务学习共享特征零样本图像分类　出处：《电子与信息学报》2017年07期 　论文类型：期刊论文

【摘要】：在利用相对属性学习实现零样本图像分类中,现有的方法并没有考虑属性与类别之间的关系,为此该文提出一种基于共享特征相对属性的零样本图像分类方法。该方法采用多任务学习的思想来共同学习类别分类器和属性分类器,获得一个低维的共享特征子空间,挖掘属性与类别之间的关系。同时,利用共享特征来学习属性排序函数,得到基于共享特征的相对属性模型,解决了相对属性学习过程中丢失属性与类别关系的问题。另外,将基于共享特征的相对属性模型用于零样本图像分类中,有效提高了零样本图像分类的识别率。实验数据集上的结果表明,该方法具有较高的相对属性学习性能和零样本图像分类精度。
[Abstract]:In the realization of zero-sample image classification using relative attribute learning, the existing methods do not consider the relationship between attributes and categories. In this paper, a zero-sample image classification method based on shared feature relative attributes is proposed, which uses the idea of multi-task learning to jointly learn class classifier and attribute classifier, and obtain a low-dimensional shared feature subspace. Mining the relationship between attributes and categories. At the same time, we use shared features to learn the sorting function of attributes, and get the model of relative attributes based on shared features, which solves the problem of losing the relationship between attributes and categories in the process of learning relative attributes. The relative attribute model based on shared feature is applied to zero sample image classification, and the recognition rate of zero sample image classification is improved effectively. This method has high relative attribute learning performance and zero sample image classification accuracy.
【作者单位】：中国科学院电子学研究所苏州研究院;中国科学技术大学软件学院;
【基金】：国家自然科学基金(41501485)~~
【分类号】：TP181;TP391.41

【相似文献】