中文摘要：

随着网络的迅速发展和应用,网络语义标签已经开始广泛地用于图像内容的标注和分享。由于图像本身及不同主体对图像的不同理解会造成图像分析理解的差异,因此如何充分利用网络标签对图像进行准确分析理解成为本文主要研究内容。图像特征包含了图像本身的大量内容信息,为了能够建立图像内容信息与网络标签之间的关系,本文主要工作包括：1,建立低层特征与图像间相似性关系;2,建立基于随机漫步模型平衡图像内容及网络标签间的关系以达到准确对图像内容分析和理解效果。实验结果表明本文所提出方法的有效性和可行性。

英文摘要：

With the rapid development and application of Web 2.0, semantic tags have been used extensively to describe the image content on the Web. As the analysis and comprehension of images may differ from person to person, we attempt to address this problem by making full use of such tags. As the low-level visual features can provide fruitful information, we propose a unified framework which includes two parts： 1） a correlation between visual similarity feature graphs and image tag bipartite graphs; 2） a random-walk model to balance the relationship between image content and semantic tags. Experiment results show that this proposed framework is effective and feasible.