一种基于SVDD的图像自动标注方法

AN SVDD-BASED AUTOMATIC IMAGE ANNOTATION METHOD

将图像自动标注看作是一个多类分类问题,目前的方法大多将其分解为两类分类问题,随着训练数据和类别的增多,这样处理时间开销将急剧增加。针对该问题,提出了一种基于支持向量数据描述SVDD(Support Vector Data Description)的标注算法,能够同时降低训练和标注过程的时间复杂度,具有易于扩充的优势。该算法利用SVDD获得包含单类数据的最小超球,通过超球边界对未标注图像进行分类并实现语义标注。在Corel数据集上的实验结果表明,该方法能显著提高图像标注的效率,并具有较好的标注性能。

In practice,automatic image annotation is deemed as the classification of multiple classes issue.At present the way commonly used is to decompose it into binary class classification problem.The time overhead of this approach increases sharply along with the growth of training data and classes.Aiming at this problem,this paper proposes an SVDD-based annotation method,it can reduce time complexity of both training and annotation process and has the advantage of facile expanding.It uses SVDD to obtain a minimum bounding hyper-sphere containing one-class data,and classifies unlabeled image via the boundary of hyper-sphere and annotates the semantics accordingly.Experimental results on Corel dataset show that the proposed method can greatly speed-up the annotation efficiency,while maintaining a competitive annotation performance.