Testing methods to estimate spectral reflectance using datasets under different illuminants

With the rapid development of mobile imaging devices, there is a strong desire for accurate image reconstruction. However, the conventional colorimetry system is often related to environmental illuminations. Spectral reflectance is the fingerprint of color and is invariant with environmental illuminations. Therefore, the goal of the present study is to verify different algorithms to reconstruct the spectral reflectance from the camera red, green, and blue responses. In this research, the weighted local sample selection method was first combined with pseudo inverse matrix method (PI) and Wiener estimation method (WE) to investigate the optimal sample number on the model accuracy. The optimum local sample numbers of the two combination methods were established. The performance of five methods was evaluated, including PI, WE, smoothing constraint method, weighted pseudo inverse matrix method (WPI) and weighted Wiener estimation method (WWE) under lightings varying a wide range of correlated color temperature (CCT) from 3000 to 10 000 K. The best algorithm (WPI) in different lighting environments was established. The metamerism of different materials was revealed, the impact of materials on training and testing samples was reported. Finally, the methods' performance under different CCTs was revealed in terms of root mean square error and CIEDE2000, and the results from the theoretical simulation and real camera capturing were compared.