In this paper we study up to what extent neural networks can be used to accurately characterize LCD displays. Using a programmable colorimeter we have taken extensive measures for a DELL Ultrasharp UP2516D to define training and testing data sets that are used, in turn, to train and validate two neural networks: one of them using tristimulus values, XYZ, as inputs and the other one color coordinates, xyY . Both networks have the same layer structure which has been experimentally determined. The errors from both models, in terms of ΔE00 color difference, are analysed from a colorimetric point of view and interpreted in order to understand how both networks have learned and how is their performance in comparison with other classical models. As we will see, the comparison is in average in favor of the proposed models but it is not better in all cases and regions of the color space.
Joan Prats-Climent, Luis Gòmez-Robledo, Rafael Huertas, Sergio García-Nieto, María José Rodríguez-Álvarez, Samuel Morillas, "A study of neural network-based LCD display characterization" in Proc. IS&T London Imaging Meeting 2021: Imaging for Deep Learning, 2021, pp 97 - 100, https://doi.org/10.2352/issn.2694-118X.2021.LIM-97