Title: Effect of Color Sample Lightness on the Performance of PCA Method in Representation of Spectral Reflectance
Abstract: Principal component analysis (PCA) has been used in color technology over 50 years. Along these years many researches have been done to find out new usages or improve its shortcomings in different applications in color technology. One of the most difficult tasks is to find suitable basis vectors, which are able to reconstruct data accurately by the lowest dimension. In this article the effect of the lightness of the data set ( L* ) on the color difference error in dimensionality reduction and recovery of the spectral reflectance using PCA method was studied. To be able to have a comparison all the experiments carried out by 3 basis vectors. The results show that by increasing the L* value of the test and training data set in dimensionality reduction using PCA, the color difference error decreases. In recovery of spectral reflectance, the L* of the training data set has no noticeable effect and the color difference error under reference illuminant is insignificant. This result can be explained by considering that, in recovery of spectral reflectance by PCA method, a metameric pair of original sample is reconstructed. Hence, the original and recovered samples almost match under reference illuminant. Although by increasing the L* of the training data set, the ΔE value under illuminant A decreases slightly, so metamerism problem can be reduced.