摘要： 利用典型相关分析(CCA)和隶属度的思想，提出一种基于模糊典型相关分析的图像特征提取新方法。通过分析图像样本的分布特点，定义合适的隶属度函数描述图像空间的样本分布。利用CCA进行多信息源特征提取，得到同时包含图像灰度信息和分布信息的有效判别特征。可证明Fisher线性判别分析是该算法的一种极限情形。在ORL标准人脸数据库上的实验结果表明新特征具有良好的分类能力，证实了该方法的有效性。

关键词: 特征提取, 典型相关分析, 隶属度, Fisher线性判别分析, 图像识别, 人脸识别

Abstract: According to the idea of canonical correlation analysis (CCA) and degree of membership, a new method of image feature extraction and recognition is proposed based on fuzzy canonical correlation analysis. An appropriate membership function is defined to express the distribution of image samples. CCA is applied to extract features from multiple information sources, which combine the information about gray level and the distribution of samples together. In addition, Fisher linear discriminant analysis is an extreme situation of the new method. The results of experiments on ORL face databases show that the features have the powerful ability of recognition, and that the presented method is efficient.

Key words: feature extraction, canonical correlation analysis (CCA), degree of membership, Fisher linear discriminant analysis (FLDA), image recognition, face recognition

中图分类号: 

• TP391.4