Trademark Retrieval Algorithm Based on Combination of Boundary and Region Features

DOI：

 作者 单位 E-mail 宋瑞霞 北方工业大学 理学院,北京 100144 songrx880@sohu.com 孙红磊 北方工业大学 理学院,北京 100144 王小春 北京林业大学 理学院,北京 100083 齐东旭 北方工业大学 理学院,北京 100144澳门科技大学 资讯学院,澳门

商标图像检索的目的是对商标图像之间的重复性、相似性进行审查.首先把商标图像的轮廓视为一个几何形状,通过一类称作V系统的正交函数系,对这个几何形状进行精确的数学表达,从而在频域求得商标的边界特征向量,这个特征向量描述了商标的整体特征.另一方面,为了描述商标的局部特征,通过对商标图像区域的划分,分别考虑各子块像素的比重和子块重心的位置,得到商标图像的两个区域特征向量,它们描述了商标的局部特征.最后利用边界和区域特征向量间的欧式距离的加权,来进行商标之间的相似度量,得到一类新的商标检索算法.大量的、各种类型的商标检索实验表明,与Fourier描述子、Zernike矩、不变矩以及Fourier描述子和Zernike矩相结合的方法等相比,方法在检索性能上有较大优势.

The purpose of trademark retrieval is to ensure that new trademarks do not repeat any images of the vast number stored in the trademark registration system. This paper regards the contour of a trademark as a geometric object and represents it precisely in terms of mathematical expressions by employing an orthogonal complete function system, V-system, as a mathematical tool. The boundary feature vector, which captures the overall features of the trademark, is calculated in the frequency domain first. Next, two region feature vectors describing the local characteristics are created by dividing the trademark image into small blocks and considering the distributions of its pixels. Finally, a new trademark retrieval algorithm is achieved by utilizing the weighted Euclidean distance between boundary and region feature vectors. The study conducts different kinds of experiments and adopts several of the evaluation criterions to evaluate the performance of the proposed algorithm. Compared with the methods of classical Fourier descriptor, Zernike moments, Hu invariant moment and combination of Fourier descriptor and Zernike moments, the proposed algorithm has obvious advantages.
HTML  下载PDF全文  查看/发表评论  下载PDF阅读器