基于形状特征的商标图像检索系统

The Shape-Based Trademark Image Retrieval System

摘要: 目的：本文采用图像形状特征来描述商标图像特征点，建立基于形状特征的商标图像检索系统，以实现更直观、准确和快捷的商标图像检索。方法：本文在Canny算法的基础上提出一种改进的自适应Canny边缘提取算法，在Shape Context的基础上提出能反映图像内部结构的Inner-Angle Shape Context，并将新算法应用于图像检索系统中。结果：设计和实现了基于形状特征的商标图像检索系统，该系统由商标预处理、特征提取和商标检索三个模块组成，实现了对任意商标在已有商标数据库中的检索，返回与样本商标最相似的15幅商标图像。结论：商标图像检索系统的实验结果表明，该系统前6个检索结果的查准率达85%，查全率达22%。

Abstract: Objective: This paper uses shape feature to describe the trademark image feature points, and establish the trademark image retrieval system based on shape feature, in order to realize more intuitive, accurate and fast trademark image retrieval. Method: This paper puts forward an improved adaptive Canny edge detection algorithm and Inner-Angle Shape Context that are applied to trade-mark image retrieval system. Results: A trademark image retrieval system is designed based on shape feature, which consists of preprocess, feature extraction, and trademark retrieval. It realizes the retrieval of any trademark in the existing trademark database, and returns the 15 most similar trademark. Conclusion: The experimental results of the trademark image retrieval system show that the accuracy rate of the first 6 retrieval results of the system is 85%, and the total recall rate is 22%.