Hybrid approach to design of the composition of automotive paint to match the desired color based on neural networks and lighting simulation
Modern automotive paints have a complex structure, and modeling their optical properties is a challenge. The inverse problem - the design of the paint composition according to its appearance - is most in demand in practical application. The shortcomings of popular mathematical methods, including previously used by the authors, are analyzed in the paper. A hybrid approach based on deep learning of a neural network and modeling of light propagation in a multilayer paint is proposed. The neural network algorithm solves the problem well for the pigments and paints on which it is trained, but is unstable for new pigments. In this case paint simulation helps to find an acceptable result. The mathematical model here provides only the functional form of the equations in variations, and the values of all functions are obtained by a few measurements which form a pigment library for future use.
lighting simulation, light scattering by particles, paints, BRDF, measurements, minimization of discrepancy, adding-doubling method
Publication language:russian, pages:17
Mathematical modelling in actual problems of science and technics