Understanding hydrogen combustion is crucial for the development of novel processes. Despite of hydrogen’s small size, hydrogen combustion is complex due to the involved diffusion effects. Although hydrogen flames are said not to emit in the VIS spectrum, literature reports significant VIS emissions of hydrogen flames. The flame region can be determined by examining the blue color channel of standard color images, as certain radical recombination reactions emit in the blue region of the spectrum. On the contrary, the red color channel can identify hot flame regions because thermally excited water vapor emits between 616 nm and 716 nm. Furthermore, H-α emissions (~656 nm) can also contribute to the red color channel. Unsupervised, physics-constrained, physically plausible spectral reconstruction (UPCPPSR) provides a cost-efficient way to reconstruct the emission spectrum of hydrogen flames and reveal the flame details from color images. METHODS Three different UPCPPSR algorithms are tested to retrieve the original VIS spectrum emitted by the hydrogen flame from RGB values. Different algorithms based on pure mathematical regularization, algebraic thermal seeding, and joint spectral-thermal optimization are investigated and evaluated on a test bench that encodes synthetic emission spectra into RGB values. The real camera spectral efficiency is used to generate the synthetic RGB values. With the ground truth known, the reconstruction algorithms can be thoroughly validated. The mean square error (MSE) and the spectral angle mapper (SAM) are used to evaluate the reconstruction quality. Unsupervised algebraic thermal seeding and joint optimization approaches are capable to reconstruct hydrogen like VIS spectra from synthetic RGB values. While the overall reconstructed spectra yield reasonable results for identifying species-specific emissions, the predicted temperatures deviate significantly from the groundtruth temperatures. Furthermore, the reconstruction quality is sensitive to the objective function and the numerical constraints of the optimization algorithms. CONCLUSIONS The results show that UPCPPSR can be a promising cost-efficient method in combustion research. However, algorithmic improvements are required before working with real flame images: i) The determination of the temperature and ii) the sensitivity to the algorithmic settings need a thorough investigation. Furthermore, the method needs to be validated with real flame emissions.