Comparative Study of Multispectral Image‐Based Auto White Balance With Optimized Conditions

Multispectral images (MSI) offer richer spectral information than conventional RGB images. Utilizing these data to predict the illuminant’s spectral power distribution (SPD) presents a potential approach for improving auto white balance (AWB). While extensive research exists for RGB‐based AWB, the optimal conditions for this MSI‐based approach have not yet been established. This study, therefore, conducts a comparative investigation to identify these optimized configurations. We compared two output methods (linear combination of standard illuminants vs. direct estimation) and two input representations (log‐chroma histogram transformation vs. direct input), across three input types (MSI, raw RGB, and MSI + RGB). Our findings demonstrate that the best model, using MSI as input, achieved the lowest color difference of 0.489, outperforming models that used log‐chroma histogram transformation (0.711) and the linear combination of standard illuminants (2.113). These results highlight the potential of MSI for enhancing AWB performance by predicting the illuminant’s SPD, while further research is required to improve robustness in real‐world applications.