The Dimension Reduction For High-dimensional Multispectral Space

The dimension reduction of high-dimensional multispectral space plays important roles in Spectral Color Management System (SCMS). They are the basis of spectral color gamut mapping technique and color gamut visualization. Firstly, In view of the Principal Component Analysis's disadvantage that the sample is non-normal and so leading to the instability of the Principal Component Analysis(PCA), a new spectral dimension reduction algorithm named ICA-Wyszecki is proposed. Combining the Independent Component Analysis and the Wyszecki hypothesis, a new low-dimensional Interim Connection Space (ICS) named LabICA is obtained. Secondly, in view of the disadvantage that the spectral reconstructed by the Interim Connection Space is beyond the spectral range because of that the Interim Connection Space obtained by Principal Component Analysis is negative. The relationship between eigenvector matrix and the ratio matrix of absorption coefficient k and scattering coefficient s can be obtained by combining the Principal Component Analysis and the Kubelka-Munk theory. In view of the disadvantage that Kubelka-Munk theory can result residual errors and Principal Component Analysis is instability, a new linear space namedψspace is proposed, and the dimension reduction forψspace is named KM-PCA algorithm. It is proved by experiment that the spectral difference,color difference and mesmerisms index between the reduced-space's description carried out by KM-PCA algorithm or ICA-Wyszecki algorithm and Spectral Profile Connection Space's description are smaller than the Principal Component Analysis's.