Synthetic data spectral rendering enhancement based on colour variation palette matrix computation

Synthetic data offer numerous advantages over real data: in terms of the amount of data a simulator is capable of generating, control over simulation conditions (such as atmospheric parameters) and, particularly in the case of supervised Machine Learning (ML), in terms of automatic labelling. However, synthetic data remain distinct from real data. In this regard, this can pose a problem for ML algorithms, as there are many unknowns regarding the features to be extracted. For example, CHORALE, a toolkit designed for 3D environments modelling, is capable of generating synthetic images in different Electro-Optical (EO) domains, such as infrared or Radio Frequency (RF). These observables are intrinsically linked to the physics of materials. In SE-Workbench (SE-WB)*, a colour in the visible spectrum is linked to a physical material, ensuring consistency in the infrared domain. By interpolating colours, the number of colours present in a scene can be increased, even from a finite list of materials. However, this poses major problems, such as rendering instability. To remedy this, a procedure for varying the colours of a material palette matrix is introduced. The principle consists in finding the pair of colours closest to the initial texture in the Red Green Blue (RGB) image using a K-Dimensional tree (KD-tree) search algorithm. This new approach makes it possible to control the visible spectral rendering of the calculated pixel colour. The paper presents this new method to enhance the realism of the pixel colour in images, both in visible and infrared spectral rendering domains.