Vol. 46, Issue 3, pp. 365-374 (2016)

Abstract

Spectral images provide richer information than colorimetric images. A high-dimensional spectral data presents a challenge for efficient spectral reconstruction. In conventional reconstruction methods it is very difficult to obtain good spectral and colorimetric accuracy simultaneously. In this paper, a segmented principal component analysis (SPCA) method and a weighted segmented principal component analysis (wSPCA) method are proposed for efficient reconstruction of spectral color information. The methods require, firstly, partitioning the complete spectrum of wavelengths into two subgroups, considering the sensitivity of human visual system. Then the classical principal component analysis (PCA) carried out each subgroup of data separately. The results indicated that the spectral and colorimetric accuracy of the SPCA and wSPCA outperformed the PCA and weighted PCA, and wSPCA clearly retained more color visual information.