Performances of Classic and Weighted Versions of Some Selected Methods in Estimation of Spectral Data from Camera Responses

The reconstruction of spectral reflectances of different color charts from the corresponding RGB values is investigated. The pseudo inverse (PI), principal component analysis (PCA) and canonical correlation regression (CCR) as well as their weighted versions are employed to estimate the spectral reflectances of Kodak Q_60, Color checker SG and Munsell data sets from the responses of a digital camera which are contaminated with different levels of random and quantization noises.The root mean square (RMS) error between the reconstructed and actual reflectances as well as the CIELAB color difference values under illuminant A for CIE1964 standard observer are computed to evaluate the performances of the employed techniques. Two different modes of camera responses i.e. three and six channels types are considered. In addition, the responses of a three-channel camera are computed under two sets of illumination conditions to prepare two collections of RGB data. To analyze the performances of the methods, they are also evaluated in the cross media condition, i.e. using different training and testing packages. According to the results, the wPI method, which is the simplest method among the other spectral reconstruction techniques, shows the greatest robustness at different levels of quantization and random noises.