Abstract: In the light of the characteristic of large-scale patent image feature database, this paper uses the method of Bounding Box-Contour Distances（BBCD features） and block features to extract the low-lever visual features, and combines the algorithm of classify index based on K-means clustering, considering the semantic similitude and the visual features similitude, to construct the index structure of data in patent image database. Thus it achieves the retrieval after classify function. The result shows, this method can increase the search speed and improve the semantic sensitivity.

Key words: feature extraction, cluster, semantic classification, image retrieval

摘要： 针对大规模专利图像特征库的特点，使用边缘轮廓距离与分块特征相结合的方法提取低层视觉特征，结合基于K均值聚类的分类索引方法，兼顾语义相似和视觉特征相似，对专利图像库数据构建索引结构，实现了先分类后检索的功能。实验结果表明，方法不仅提高了检索速度，而且提高了检索的语义敏感度。

关键词: 特征提取, 聚类, 语义分类, 图像检索