ICA和改进的SVM在有限集字符识别中的应用  

Application of Independent Component Analysis and Improved Support Vector Machines on Finite Set Character Recognition

作　　者：鹿晓亮[1] 陈继荣[1] 黄戈祥[1] 

机构地区：[1]中国科学技术大学电子工程与信息科学系,合肥230027

出　　处：《计算机工程》2006年第23期199-201,共3页Computer Engineering

摘　　要：介绍了独立分量分析(ICA)基本原理和算法,提出了一种基于独立分量分析和支持向量机的有限集字符识别新方法。对传统向量机解决多分类问题的“一对一”模式进行了改进,将传统向量机的“一对一”模式存在的不可分区域减小到可以忽略的程度,克服了不可分区域的影响。该算法可应用于车牌字符、手写体英文字母、手写体数字、印刷体字母、印刷体数字等有限集字符的识别。在大量的车牌汉字和手写体英文字母自动识别实验中,取得了高于95%的识别结果,证明该算法在有限集字符识别应用中的优越性。This paper introduces the principle and algorithm of independent component analysis （ICA） and then puts forward a new method to recognize finite set characters utilizing ICA and support vector machines（SVM）. It develops the ＂one-versus-one＂ model of multi-classification of traditional SVM and at the same time reduces the influence of undivided area. The presented algorithm can be applied to the recognition of license plate characters, handwritten English letters, handwritten numbers, printed letters, printed numbers and other finite set charaeters. The experiments to recognize Chinese characters on license plates and handwritten English characters can achieve a recognition rate of more than 95%. The results show that this algorithm holds advantage in the finite set characters recognition.

关 键 词：独立分量分析 支持向量机 特征提取 字符识别 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]