摘要
为提高商标检索的查准率和快速性,提出一种将全局特征修正HU不变矩和局部特征SIFT相结合的MHU-SIFT集合型商标检索方法。该方法首先对图像进行初步的滤波去噪等处理,其次,提取图像的HU修正矩和角点特征SIFT建立索引,最后运用基于图像内容的检索方法(CBIR)对商标进行检索。在机器为2GB+64位的笔记本电脑,使用标准2000张商标图像库进行检验。结果表明,MHU-SIFT方法查准率较HU-SIFT方法提高了10%。
In order to improve the precision and rapidity of trademark retrieval, a method of trademark retrieval algorithm is proposed by global features modified HU moment invariants and local features SIFT, so as to MHU-SIFT. This method firstly carries on preliminary filter of image denoising processing. Second, the image of modified HU moment invariants and SIFT features is extracted. And then the index is built by using the method. Finally, trade- mark is retrieved by using the method based on image content retrieval（CBIR） method. The proposed retrieval technique is tested using the standard trademark database of 2000 images. Simulations results show 10% precision/recall improvement in the case of trademarks data- base, in the machine for 2GB ＋ 64 Laptop.
出处
《沈阳理工大学学报》
CAS
2017年第6期78-83,93,共7页
Journal of Shenyang Ligong University
基金
辽宁省教育厅科学技术研究项目(L2014083
L2015467)
辽宁省自然科学基金指导计划项目(201602651)
辽宁省重点实验室开放基金项目(4771004kfs55)
关键词
商标检索
全局特征
局部特征
HU修正不变矩
尺度不变特征变换(SIFT)
trademark image retrieval
global feature
local feature
HU modified invariantmoments
scale-invariant feature transform （SIFT）
