摘要： 将因子化主成分分析(FPCA)算法应用于人脸图像特征提取时，需要使用迭代算法，但该算法应用于高分辨率图像时实时性较差，并且可能导致维数灾难。针对上述问题，提出一种模块化FPCA(M-FPCA)的新型特征提取方法。将原始数字图像样本进行模块化，对模块化后得到的各个子图像矩阵采用FPCA 算法进行特征提取，合并子图像特征矩阵得到原图的特征矩阵。彩色图像由R、G、B 3个分量来表示，根据现有彩色信息融合方法的不足，对其进行改进，并结合M-FPCA算法，提出一种彩色M-FPCA新方法。在CVL和FEI人脸库上进行的实验结果表明，M-FPCA方法能提高FPCA算法的实时性，解决维数灾难问题。彩色 M-FPCA方法能有效提取彩色人脸图像的色彩信息，得到较高的人脸识别率。
关键词:
主成分分析,
因子化主成分分析,
模块化FPCA,
彩色M-FPCA,
特征提取,
彩色图像识别
Abstract: Using Factored Principal Components Analysis(FPCA) feature extraction algorithm in the high resolution images has bad real-time performance and may cause dimension disaster because it needs to use iterative algorithm to implement FPCA algorithm. In order to solve the problem above, this paper developes a new method called Modular-FPCA(M-FPCA) for image feature extraction. This method modularizes the original digital image samples, implements FPCA algorithm on every sub-image matrix, and gets feature matrix of original image by merging sub-image features. Color images can be represented by three components of R, G, B. For existing shortcomings of color information fusion method, it combines M-FPCA with the improved color information fusion method and names it as color M-FPCA. Experimental results on CVL, FEI color face image library show that M-FPCA method can improve the realation of FPCA algorithm, solve dimesion disaster problems, color M-FPCA method extracts color information from color face image effectively, and has higher recognition rate.
Key words:
Principal Components Analysis(PCA),
Factored PCA(FPCA),
Modular-FPCA(M-FPCA),
color M-FPCA,
feature extraction,
color image recognition
中图分类号: