融合小波变换和SIFT特征的商标检索方法  

Trademark Retrieval Method Combining Wavelet Transform and SIFT Features

作　　者：汪慧兰[1] 毛晓辉[1] 杨晶晶[1] 杨卫中[1] 

机构地区：[1]安徽师范大学物理与电子信息学院,安徽芜湖241000

出　　处：《计算机技术与发展》2015年第4期89-92,共4页Computer Technology and Development

基　　金：安徽省高校省级自然科学研究项目(KJ2011Z135);中科院光电技术研究所微细加工光学技术国家重点实验室开放基金

摘　　要：针对基于SIFT特征的商标检索其计算量大、检索速度慢、效率低的不足,提出了一种融合小波变换和SIFT特征的商标检索方法。该方法先通过小波分解得到低频子图像并建立低频子图像库,在低频子图像上提取商标图像的SIFT特征,再逐一进行精确匹配,对相似度进行排序,输出检索结果。实验结果表明,该方法不仅保持了SIFT特征的良好描述能力,且减少了精确匹配的计算量,提高了匹配速度。融合小波变换和SIFT特征的商标检索方法有效地弥补了原算法的不足,表现出了更好的检索能力和性能。Trademark search based on SIFT has some shortcomings, such as the large amount of calculation, slow search speed, low efficiency and so on. A method for trademark search by integration of wavelet transform and SIFT features was proposed. First, get low-frequency sub-image by wavelet decomposition, and create a low-frequency sub-image database. Then, extract SIFT feature in the low-frequency sub-image after wavelet transform, do exact match one by one, and be sort of similarity, thus output search results. The experimental results show that this method not only maintains a good description of SIFT features, but also reduces the amount of calculation of precisely matching and improves matching speed. The trademark retrieval method combining wavelet transform and SIFF features can effectively compensate for the deficiencies of the original algorithm,and show better search capabilities and performance.

关 键 词：商标检索 小波变换 SIFT 特征匹配 检索性能 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]