Abstract: In order to overcome the semantic gap between low-level features and high-level semantic concept of image, an image annotation approach based on visual attention weight and word correlation was proposed. In the process of understanding an image, people pay more attention to focus region. Focus region of an image can be extracted by computing visual attention weight of image regions. The annotation word of focus region is relevant to the annotation word of other image regions, so we can choose proper annotation vector by word correlation. The experimental results show that the new method can improve the precision of image annotation.

Key words: semantic gap, visual attention, WordNet, word correlation, image annotation

摘要： 为了弥补图像底层特征到高层语义之间的语义鸿沟，提出一种基于视觉焦点权重模型和词相关性的图像标注方法。由于人们对图像的认识过程中，对焦点区域有比较多的关注，因此可以通过视觉焦点权重模型计算图像各区域的视觉焦点权重来提取图像的焦点区域。同时焦点区域的标注词和其他区域的标注词在逻辑上是相关的，因此通过WordNet根据词汇相关性确定图像的最终标注向量。实验结果表明，通过该方法能提高图像自动语义标注的准确率。

关键词: 语义鸿沟, 视觉焦点, WordNet, 词相关性, 图像标注