基于回归相关模型的自动图像标注

Regression Based Relevance Model for Automatic Image Annotation

如何挖掘基于语义的相关模型是当前自动图像标注技术中一项重要而迫切的研究课题。本文从相关概率模型的角度分析了回归技术解决自动图像标注任务的可行性,进而提出了基于回归相关模型(Regression RelevanceModel,RRM)的自动图像标注框架。RRM通过有效地建立了图像-关键词的相关性,准确地描述了图像视觉特征与语义关键词之间的概率关系,从而实现图像的语义标注任务。最后在COREL数据集上的实验,表明基于RRM图像标注方法的有效性。

A popular technology is focused on how to build the semantic relevance model for the task of automatic image annotation. According to the probabilistic relevance model, it is feasible and effective for regression analysis to deal with the problem of image annotation. Thus, regression based relevance model （RRM） is proposed to implement the image annotation in this paper. The relation of image - keyword is effectively constructed via RRM, which correctly characterizes the probabilistic correlation between the visual feature of images and the semantic keyword. The experimental results on COREL reveal that high annotation accuracy can be achieved by the proposed framework.