Interim Connection Space Based on Polynomial Models for Multispectral Image Compression and Reconstruction

In order to improve the accuracy of multispectral image reconstruction, this paper proposes an interim connection space called PMLabPQR, which is an improvement of LabPQR. The first three dimensions of this space are the three-dimensional chromaticity information of a given spectrum S in a specified light environment. Compared with LabPQR, which uses the "R matrix" as the mapping matrix for converting the triple-stimulus values to the estimated spectra, PMLabPQR establishes the mapping relationship between the triple-stimulus values of CIEXYZ and the original spectra through polynomial regression, and then obtains the coefficients of the polynomial model by the least-squares method to obtain the coefficients of the polynomial model. The coefficient matrix of the polynomial model was obtained by the least-squares method as the mapping matrix, and the error between the estimated spectra and the original spectra obtained from the mapping matrix was analyzed by principal component analysis (PCA) to obtain the first three principal components as the last three dimensions of interim connection space. The reconstructed spectra are composed of the estimated spectra from the mapping matrix and the polynomial model consisting of the three stimulus values, and the spectral errors obtained by doing the inverse PCA variations. The experimental statistical results show that the new method is significantly better than LabPQR, MlabPQR, and tD65A in spectral reconstruction accuracy, and slightly lower than tD65A in chromaticity reconstruction accuracy, but with reduced additional information. Therefore the new method can be better applied to the compression and reconstruction of multispectral images.