基于Zernike矩的字符特征归一化方法  

A method to normalize character feature based Zernike moment

作　　者：李了了[1] 邓善熙[2] 丁兴号[3] 

机构地区：[1]内江师范学院计算机信息科学系,四川内江641112 [2]合肥工业大学仪器科学与光电工程学院,安徽合肥230009 [3]厦门大学电子工程系,福建厦门361005

出　　处：《微计算机信息》2007年第04X期260-261,263,共3页Control & Automation

基　　金：安徽省"十五"科技攻关重点资助项目(01022010)

摘　　要：提取Zernike矩的幅值作为字符的模式特征具有旋转不变性,但单位圆半径的选择对特征计算影响较大。本文提出一种根据字符外部边框信息计算最佳单位圆半径的归一化方法。经实验证明该方法能保证单位圆合适地把整个字符包含在内,同时由于单位圆半径不变使数字图像离散化对Zernike矩旋转不变性的影响降到最低。The character＇s pattern feature extracted from Zernike moment amplitude has rotation invariance, but the choice of unit circle radius affect the feature directly. The article presents a normalization method that can obtain the best unit circle radius based on information of character＇s contour. Experiments demonstrated that the radius could cover the whole character properly by this method, at same time, because of radius invariance the influence of digital image＇s discretization on rotation invariance of Zernike moment was reduced to the smallest level.

关 键 词：ZERNIKE矩 字符特征 归一化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]