Multispectral compression and reconstruction using weighted PCA with consideration of color difference caused by tiny wavelength change

• Published in: Optical review (Tokyo, Japan) Japan), 2023-06, Vol.30 (3), p.275-289
• Main Authors: Cao, Qian, Cui, Qingbin, Ge, Jinghuan
• Format: Article
• Language: English
• Subjects: Atomic; Lasers; Microwaves; Molecular; Optical and Plasma Physics; Optical Devices; Optics; Photonics; Physics; Physics and Astronomy; Quantum Optics; Regular Paper; RF and Optical Engineering
• ISSN: 1340-6000; 1349-9432
• DOI: 10.1007/s10043-023-00807-x

To retain more color information in multispectral compression and reconstruction for spectral color reproduction, a weighted principal component analysis with consideration of color difference caused by tiny wavelength is proposed in this paper. The weight function, which considers the final tool for evaluating multispectral compression and the reconstruction algorithms is color difference, is the average value of spectral color differences between the spectra of a spectral dataset and the spectra obtained by subtracting tiny value from the spectral dataset. Spectral color difference formula is introduced to calculate spectral color difference between the two spectra. NCS, Munsell, and SOCS (ISO/TR 16,066:2003) are used to construct three weight functions, SCDWF-1, SCDWF-2, and SCDWF-3, respectively, to obtain the corresponding weighted principal component analysis, SCDPCA-1, SCDPCA-2, and SCDPCA-3. The root mean squared error ( RMSE ) and goodness fitting coefficient ( GFC ) are employed as the spectral evaluation index and the CIELAB color difference is employed as the colorimetric evaluation index. The feasibility and performance of the proposed methods are tested by comparing the results of principal component analysis (PCA) and the other two weighted PCA by compressing and reconstructing three different sets of test samples NCS, Munsell, and SOCS. Statistical results show that compared with PCA, the proposed SCDPCA can significantly improve the colorimetric accuracy at the expense of a small amount of spectral accuracy. Moreover, the colorimetric and spectral accuracy of SCDPCA is better than that of the other two weighted PCA recently proposed by other researchers.