光谱重建中ICA和PCA的对比  

Comparison of ICA and PCA in Spectrum Reconstuction

作　　者：张小燕[1] 杨卫平[1] 冯洁[1] 李宏宁 孙波[1] 杨晓莉[1] 段剑金[1] 石俊生[1] 

机构地区：[1]云南师范大学物理与电子信息学院

出　　处：《云南师范大学学报（自然科学版）》2007年第3期28-35,共8页Journal of Yunnan Normal University:Natural Sciences Edition

基　　金：国家自然科学基金(60368001);云南省自然科学基金(2005F0033M)

摘　　要：一个光谱数据集可以表示成几个主要光谱成分的线性组合。主成分分析法(PCA)是提取光谱数据集的主要成分的常用方法。近年来,有研究人员用独立成分分析法(ICA)提取光谱数据集的独立成分,进而实现光谱数据压缩。文章分别使用ICA和PCA对50例Munsell色卡的光谱反射比和50例桦树叶的光谱反射比进行特征光谱的提取。利用多光谱成像技术和光谱重建算法,采用三组滤光片,每组分别为2、3和4片,选取3-15维子空间,重建了150例Munsell色卡和150例桦树叶的光谱反射比。重建结果用CIELAB1976色差和光谱重建误差来评价。在150例桦树叶光谱重建中,ICA的重建结果明显好于PCA的重建结果;而150例Munsell色卡用ICA和PCA重建结果相差不大。最后,根据重建结果,对子空间维数、滤光片数与重建色差和误差的关系作了分析。A dataset of color spectra can be represented as a linear combination of few principal spec-tra. The principal components of a spectral dataset are usually generated by the Principal Component Analysis （PCA）. The Independent Component Analysis （ICA） has been used to abstract the independent components of the spectral dataset by some researchers in recent years and enables data compression. In this paper, the feature spectra of spectral reflectance of 50 cases Munsell color cards and 50 cases birch leaves are abstracted by ICA and PCA separately. The reflectance of 150 cases Munsell color cards and 150 cases birch leaves are reconstructed with the multi - spectral imaging and the algorithm of spectral reconstruction, and with three sets of tilters, each compose of two, three, and four filters, and three to fifteen dimensions subspace. The reconstruction results are evaluated by the CIE1976 color difference and the spectrum reconstruction errors. In the reconstruction of the 150 cases birch leaves, the reconstruction results by ICA are obviously better than that of the PCA ; the differences between the results of ICA and PCA are not so significant in the reconstruction of the 150 cases Munsell color cards. Finally, according to the results of reconstruction, the relationship between the subspace dimensions, the number of filters and the reconstruction color differences, spectral error are analyzed in this paper.

关 键 词：主成分分析 独立成分分析 光谱重建 

分 类 号：O432.3[机械工程—光学工程]