Abstract: In order to better achieve the image retrieval based on semantics, the integrated features of color, texture and shape were used to represent the image and were also regarded as input vectors of Support Vector Machine (SVM). Through making study of image classes, the correlation from image low-level features to high-level semantics was built. The classification accuracy was improved by using comprehensive features. Then image library was organized by the semantic structure, and hierarchical representation of image semantics was realized. All keywords of different levels were combined to describe the semantic of images. The results show that the proposed method can make the image expressed by more comprehensive semantic in the case of getting good classification accuracy.

Key words: image classification, semantic, image retrieval, Support Vector Machine (SVM), hierarchy

摘要： 为了更好地实现基于语义的图像检索，结合了颜色、纹理和形状的综合特征来表示图像，将它们作为支持向量机(SVM)的输入向量，对图像类进行学习，建立图像底层特征和高层语义的关联。采用综合特征表示图像，提高了分类正确率。同时按照分语义层次的方式组织图像库，实现图像的语义分层表示，用各层次的关键词来联合表示图像的语义信息。结果表明，可以在具有较好分类正确率的情况下，使图像具有更全面的语义表示。

关键词: 图像分类, 语义, 图像检索, 支持向量机, 层次