摘要
利用典型相关分析的思想,提出了一种基于特征级融合的组合特征抽取新方法.首先,探讨了将典型分析用于模式识别的理论构架,给出了其合理的描述.即先抽取同一模式的两组特征矢量,建立描述两组特征矢量之间相关性的判据准则函数,然后依此准则求取两组典型投影矢量集,通过给定的特征融合策略抽取组合的典型相关特征并用于分类.其次,解决了当两组特征矢量构成的总体协方差矩阵奇异时,典型投影矢量集的求解问题,使之适合于高维小样本的情形,推广了典型相关分析的适用范围.最后,从理论上进一步剖析了该方法之所以能有效地用于识别的内在本质.该方法巧妙地将两组特征矢量之间的相关性特征作为有效判别信息,既达到了信息融合之目的,又消除了特征之间的信息冗余,为两组特征融合用于分类识别提出了新的思路.在肯考迪亚大学CENPARMI手写体阿拉伯数字数据库和FERET人脸图像数据库上的实验结果证实了该方法的有效性和稳定性,而且识别结果优于已有的特征融合方法及基于单一特征进行识别的方法.
In this paper, based on feature fusion, a new method of feature extraction is proposed according to the idea of canonical correlation analysis. At first, the framework of canonical correlation analysis(CCA) used in pattern recognition is discussed and its reasonable description is given. This comprises three steps: extracting two sets of feature vectors with the same pattern and establishing the correlation criterion function between the two sets of feature vectors; solving the two sets canonical projective vectors and extracting their canonical correlation features by the CCA algorithm; doing feature fusion for classification by using proposed strategy. Then, the problem of canonical projection vectors is solved when two covariance matrices of training samples are singular. This method is adapted to small sample size and high-dimensional problems, so the applicable range of CCA is extended in theory. At last, the inherent essence of this method used in recognition is analyzed further in theory. The proposed method uses correlation features of two groups of feature vectors as effective discriminant information, so it not only is suitable for information fusion, but also eliminates the redundant information within the features. This is a new way to classification and recognition. The experiment results on the CENPARMI handwritten Arabic numerals database of Concordia University and FERET face image database show that recognition rate is far higher than that of the algorithm adopting the single feature or the existing fusion algorithm, and that this algorithm is efficient and robust.
出处
《计算机学报》
EI
CSCD
北大核心
2005年第9期1524-1533,共10页
Chinese Journal of Computers
基金
香港特区政府研究资助局(CUHK/4185/00E)资助
关键词
典型相关分析
特征融合
特征抽取
手写体字符识别
人脸识别
canonical correlation analysis
feature fusion
feature extraction
handwritten character recognition
face recognition