Multilayer Perceptron to Boost Colorimetric Capacities of 2D RGB and Hyperspectral Snapshot Cameras

The ability to accurately measure the surface colour has enormous applications in the quality control of a wide range of manufacturing processes. Currently, the industry still relies on traditional colourimeters to implement these tasks.

In this work, we compare the behaviour of two types of snapshot cameras (RGB and hyperspectral) in the frame of accurate colour measuring. We use two approaches for colour and spectral reconstruction, one based on linear modelling and another novel one based on Multilayer Perceptron (MLP). Experiments use charts with a large number of different colours (165 for training and 165 for testing). The results showed promising properties of neural networks in this task. We reach an average 1.6 CIEDE 2000 colour difference between estimations and ground truth.