Reconstruction of spectral color information using weighted principal component analysis

Spectral reflectance is widely useful in many different applications nowadays. Accurate spectral images are usually high-dimensional data and larger files, so a proper dimensional reduction method can reduce storage space on the premise of a minimum loss of information. In this study, a nonlinear weighted component analysis (wPCA) method, considering the more optimal match of human color visual, was applied to recover reflectance. Our main aim is to retain more color information and achieve better color reproduction performance. The feasibility and performance of the wPCA was tested by compressing and reconstructing the ColorChecker 24, ColorChecker SG, Munsell and three multi-spectral images, comparing the results with the standard PCA. As results presented that the wPCA method has the characteristic of universality, stability and robustness, and clearly improves the color reproduction accuracy comparing to the PCA.