A Study on Content-Based Vector Image Retrieval Systems

本研究的目的是开发和评估一个针对矢量图像的相似性检索系统。矢量图像用于表示具有清晰表示轮廓的插图，例如商标和动画人物。该系统从数据库中的每一幅矢量图像中提取特征点，然后通过匹配特征点的空间分布来评估一对矢量图像之间的相似性。通过使用包含约1100个商标的数据库，对检索精度进行了测量。实验结果表明，该系统的检索精度优于传统的基于栅格图像检索方法的系统。

The goal of this study is to develop and evaluate a system for similarity retrieval for vector images. Vector images are used for representing illustrations which have contours clearly represented such as trademarks and anime characters. The proposed system extracts feature points from each of vector images in a database, and then, evaluates similarity between a pair of vector images by matching the spatial distributions of the feature points. By using a database containing about 1,100 trademarks, the retrieval precision has been measured. The experimental results have shown that the retrieval precision of the proposed system is superior to a conventional system based on a raster image retrieval method.