Reconstruction Algorithm from an Interim Connection Space to Spectra Based on BP Neural Network

An interim connection space (ICS) composed of two sets of tristimulus values under two different illuminants (D50, D65) is proposed and a reconstruction algorithm based on BP neural network to convert the ICS into spectra. In order to test and compare the performance of the proposed ICS with the other four ICS, the NCS spectral dataset is used as training samples, and NCS and Munsell spectral datasets are used as testing samples. The root mean square error (RMSE) between the original and reconstructed spectra is used to evaluate the spectral accuracy. The average values of mean color difference between the original and reconstructed spectra under various lighting conditions is used to evaluate the colorimetric accuracy. The experimental results show that the proposed ICS has the best performance in spectral accuracy and colorimetric accuracy compared to the other ICS, and has obvious advantages compared to the other four ICS.