&lt;title&gt;Comparison of PCA and ICA in color recognition&lt;/title&gt;

Section: Spaces Described By Pn Basesmentioning

confidence: 97%

Spectral spaces and color spaces

Ramanath

Kuehni

Snyder

et al. 2003

“…Similar to some works in the field of spectral processing, the radiation data were not mean-centered in this research. 13 Clearly in most color matching application, the importance of different lighting conditions is not identical and they can be ordered according to their weight. In the classical colorimetric approach, the matching occurs only under specific illuminant and the others are ignored.…”

Section: Pca and Wpcamentioning

confidence: 99%

Design of virtual illuminants to control the colors under multiple illuminants

Agahian

Amirshahi

2009

A series of virtual illuminants are generated by applying weighted principal component analysis (wPCA) as well as weighted mean (wM) techniques to a set of artificial lights. The weights are selected in accordance to the importance of desired light source in control and/or matching process. In fact, the created virtual illuminants can be implemented in a colorimetric color matching trial or color control processing. In the case of color matching effort, by choosing suitable weights for different lighting conditions, more balanced color difference values under those illuminants which are presented in dataset could be expected.

“…Their differences are described elsewhere. [17][18][19] Two datasets were used. The first was the 1,269 reflectance spectra of the chips in the Munsell Book of ColorMatte Finish Collection measured at the University of Joensuu, Finland.…”

Section: Different Process Primariesmentioning

confidence: 99%

Comparison of methods of parameric correction for evaluating metamerism

Zhao-jian

Berns

2007

In practice, most metameric pairs do not achieve colorimetric equality for a set of reference conditions. These parameric pairs are composed of a residual color difference and a metameric difference. Three techniques have been used to correct this residual color difference: an additive correction in L*a*b*, a multiplicative correction in XYZ (recommended by the CIE), and parameric decomposition where the batch's spectrum is adjusted. Parameric decomposition can be viewed as batch correction using three ''colorants'' (process primaries) where the color-mixing model is linear in reflectance. Most often, Cohen and Kappauf's Matrix R technique is used where the primaries are color matching functions. Alternative primaries were derived from a Munsell Book of Color and an automotive paint system using principal component analysis (PCA) and independent component analysis (ICA). 1,152 parameric pairs about 24 color centers were synthesized using the paint system and Kubelka-Munk turbid-media theory. Each parameric pair was corrected to a metameric pair using these methods. Spectral accuracy was evaluated by comparing the corrected spectra to metameric reference spectra calculated using Kubelka-Munk batch correction. The Matrix R technique had the worst spectral accuracy under the reference conditions while both PCA and ICA had similar and reasonable accuracy. The special index of metamerism, change in illuminant, was calculated for each parameric pair using these various correction techniques to achieve colorimetric equality. The Matrix R and CIE-recommended multiplicative techniques were statistically significantly worse than parameric decomposition using Munsell Book of Color PCA and ICA and automotive ICA process primaries.