Abstract: Content based trademark retrieval algorithms use overall shape features or spatial relationships between components to match with features stored in database. However they’re inadequate in anti-noise and always suffer pretermission or false matches when handling trademarks with similar components. This paper proposes a new retrieval method using the Sift descriptor. Experimental results show that this method is more accurate, more complete in retrieving similar trademarks and is more robust against noise compared with existing methods.

Key words: trademark retrieval, Content Based Image Retrieval(CBIR), Sift, Recall-Precision criterion

摘要： 基于内容的商标检索利用全局形状特征和子图像的空间位置关系进行特征匹配，存在检索精度不高、抗噪声能力不强和相似子图像可能发生误检、漏检等不足。该文提出一种利用Sift特征进行商标检索的新方法。实验结果表明，与现有检索方式相比，该方法准确全面，并对遮挡、扭曲、噪声等干扰具有更强的抵抗力。

关键词: 商标检索, 基于内容的图像检索, Sift算法, 查准率-查全率

CLC Number: 

• TP311