一种基于手绘图的商标检索算法

A Trademark Retrieval Algorithm Based on Sketch

为了高效的进行商标设计,提出了基于内容的商标检索算法.首先应用Zernike矩描述图像的形状信息,并将手绘图作为查询条件进行检索,依据形状距离进行排序,将排序结果作为第一轮检索结果输出.其次,用户对第一轮检索结果进行标注,并将标注结果反馈给系统,通过分类器学习后,再次进行检索并输出最终检索结果.实验结果表明,提出的算法在商标图像的检索中具有检索准确率高、耗时少的特点;此外,算法可支持用户以手绘的方式进行查询,有一定的实用价值.

A trademark retrieval approach is proposed to designing trademarks effectively. Firstly, Zernike moments are applied to describe the shape information of images, then sketch is used as query to search. The results are sorted based on the shape distance and shown as the 1st-round retrieval result. Secondly, the retrieval result is marked by the users, and then the marked trademarks are feedbacked to the system. Based on classifier learning, retrieve and output search results are shown again. The simulation results show that the proposed algorithm is charaterized by high accuracy and less consuming -time in the trademark image retrieval system. Besides, the proposed algorithm supports users to query by hand sketching. The algorithm is of practical value.