We consider the problem of spectral reconstruction from multispectral images by using non-linear methods. In the search for a neural network able to provide noise resistance and good generalization we apply Mixture Density Networks. This approach has been tested and compared with a linear method already used for spectral reconstruction of fine art paintings. This has been done using simulated and real data. Mixture Density Network based methods provide very good results in both cases. In particular, for real data acquisition we have scanned a Gretag-Macbeth™ color chart using a Minolta CS-100 spectroradiometer and a PCO SensiCam 370 KL monochrome camera with an electronically tunable liquid crystal spectral filter VariSpec VIS2. The results obtained using the data from this experiment clearly show the superiority of the Mixture Density Network based approach over the linear one used as a reference.
Alejandro Ribés, Francis Schmit, "Reconstructing Spectral Reflectances with Mixture Density Networks" in Proc. IS&T CGIV 2002 First European Conf. on Colour in Graphics, Imaging, and Vision, 2002, pp 486 - 491, https://doi.org/10.2352/CGIV.2002.1.1.art00102