一种融合两种主成分分析的货币识别方法  

Currency recognition based on fusion two different principal component analysis

作　　者：贺建飚[1] 崔艳丽[1] 

机构地区：[1]中南大学信息科学与工程学院,长沙410083

出　　处：《计算机工程与应用》2009年第23期170-172,共3页Computer Engineering and Applications

基　　金：国家自然科学基金No.60503007;湖南省自然科学基金No.07JJ5077~~

摘　　要：为了充分提取货币图像的有效特征,提出了一种新颖的货币识别算法。首先,利用两种主成分分析分别降低原始图像空间的维数;然后,对所得特征参数属性用粗糙集决策表约简原理进行优化;最后,运用典型相关分析原理融合特征级数据,并给出最终识别效果。实验结果表明,该算法在充分提取货币图像特征的基础上,能很好地融合其特征向量,识别性能优于单一的主成分分析方法,且训练样本为20时,识别率可达98.78%。In order to extract the most useful features of currency,a novel method of currency recognition is proposed.First of all,two kinds of Principle Component Analysis （PCA） are used to respectively reduce the dimensionalities of the original image vector space.Then,rough set attribute optimization is introduced to optimize the parameters of feature vectors.At last,canonical correlation analysis is adopted to fuse these features.The resuhs show that the performance of the proposed method can extract the more useful features,and the recognition results are better than those of using one kind of PCA.In addition,when the number of the training set is 20,the recognition rate is 98.78%.

关 键 词：货币识别 主成分分析 粗糙集 典型相关分析 特征级融合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]