基于多特征标签相关性学习的图像自动标注

Image Annotation Based on Multiple Feature Tag Relevance Learning

在网络真实环境下的图像数据集上的大规模语义标注是一个研究难点。提出了一种基于多特征标签相关性学习的图像语义标注方法,针对真实环境下大规模图像集合进行自动标注。首先提取图像多种视觉特征,采用多标记学习方法在特定特征空间完成标注词相关性学习,得到每幅图像的单特征标注词相关度;然后采用一种动态阈值确定方法估计单个特征和标注词的相关度阈值;最终采用一种无监督组合方法融合多种特征标和标注词的相关性生成图像语义标签。通过互联网数据集上的测试表明了方法的有效性。

An image annotation method was proposed based on multiple feature tag relevance learning （MFTRL）, which aimed at tagging large-scale image collections in real environment by analyzing the correlation between tags and images represented by multiple visual features. First, a multiple label learning method was utilized to generate the relevance of tags and images in specific feature space. Then, an optimal threshold was set for each tag and corresponding single feature. So the output of many tag relevance learners dlriven by diverse features could be combined in the manner of combining multi-feature tag relevance. The experiments over the intemet image set demonstrate that the proposed method is accurate and stable.