ICA与PCA特征抽取能力的比较分析

Comparison and Analysis on ICA & PCA’s Ability in Feature Extraction

独立分量分析和主分量分析在模式识别领域中,处理的方法和思路有很多相似之处。本文对两者的原理及特点进行了分析,并分别设计了两者的重建模型字符识别方法。通过对美国国家邮政局USPS字库中全部数字字符完整的识别实验,对两种方法在模式识别中特征抽取能力作了全面比较和系统分析。

In the domain of pattern recognition, there are similarities between Independent Component Analysis (ICA) and Principle Component Analysis (PCA) in their ways of conducting analysis. Firstly the theory and characteristics of ICA and PCA are analyzed in this paper. Then classification algorithms based on reconstruction models using ICA and PCA to conduct written character recognition are proposed. Finally, by using the USPS database, comprehensive comparison and ana1ysis of ICA and PCA in term of their ability in feature extraction, model reconstruction and classification are reported.