中文摘要：

图像通常用关键词表示其语义信息，基于关键词的图像检索方式存在因用户理解差异而导致对图像语义理解的歧义问题。文中利用语言学本体WordNet中单词的语义层次，并综合考虑单词之间的同义关系、上下位关系等不同层次的抽象语义信息，通过建立检索词和标注词间的语义关联，统一不同用户对图像语义的理解和描述，再结合单词在不同抽象层次的语义信息计算图像的相似性距离，实现了基于高层语义的图像检索。实验结果表明，上述方法能有效提高图像的检索性能。

英文摘要：

Aim. Words in image label frequently have multi-senses, leading to unsuitable search results for users. We now propose using WordNet to disambiguate image senses. In section 1 of the full paper, we discuss semantic hierarchy of images. Section 2 explains the calculation of image semantic similarity measure. What we do in section 2 is essentially that, we, employing the multi-relations between words, such as synonyms, hyponyms and hypernyms provided by WordNet, build the relationships between image labels, thus unifying the description of image label. The image semantic similarity measure can be calculated by eq. （7）. In section 3, we give experimental results shown in Fig. 1 and Table 1, which indicate preliminarily that the calculated image semantic similarity measure in our semantics-based image retrieval method gives, in general, better performance as compared with that of the label-based method.