图学习的区域图像标注方法

Region-based image annotation of graph learning approach

近年来,图像标注技术得到广泛关注。提出一种图学习的自动图像标注方法,将图像标注作为多示例学习框架下的半监督学习策略,通过给出适合图像在包空间的有效度量方式,充分利用未标注样本挖掘图像特征的内在规律性,将半监督学习的方法和多示例学习有效结合起来,从而获得更准确的标注结果。实验结果表明,提出的标注方法可行,同时标注结果与传统的标注方法相比得到了明显提高。

Image annotation has been an active research topic in recent years. The authors formulated image annotation as a semi-supervised learning problem under multi-instance learning framework. A novel graph-based semi-supervised learning approach to image annotation, using multiple instances, was presented, which extended the conventional semi-supervised learning to multi-instance setting by introducing the adaptive geometric relationship between two bags of instances. The experimental results show that this approach outperforms other traditional methods and is effective for image annotation.