A Neural Network Approach to Estimating Color Reflectance with Product Independent Models

In the paint and coatings industry, traditionally color reflectance modelling is performed individually for each coating product. This is because a coating product contains color samples that are mixed from several colorants and a binder which have a unique chemical property that requires modelling to be carried out individually when done analytically. This work proposes a superior approach for color reflectance modelling based on Neural Networks, which is capable of modelling multiple coating products concurrently using a single model, allowing for a modelling approach that is generic and independent of the coating products. In this study we demonstrate that our Neural Network model optimized to predict color reflectance for multiple coating products using a dataset with 4150 color samples containing 18 distinct coating products, is able to perform better (RMSE 3.73) than an widely employed analytical model, Kubelka-Munk (RMSE 8.24), which is conventionally used for the same task.