Improving colour rendering prediction with hyperspectral images

Jost Sophie – ENTPE France

The recent developments in spectral imaging offer the opportunity to obtain good quality images with accurate knowledge of the spectral content of each pixel. Indeed, today, high-quality hyperspectral cameras can capture spectral information with a spectral resolution equivalent to single-point spectroradiometers and a spatial resolution producing up to 2K images.
From the radiances of the acquired scene, spectral reflectances can be calculated and the spectral power distribution of any light source can be applied. Thus, hyperspectral images enable to simulate real complex environments under different real or virtual illuminants. Beside the opportunity to visually test colour rendering with an image (which still has the disadvantage of metamerism due to the primaries of the display device), hyperspectral image is a useful tool to investigate colour rendering in real complex environments using their specific real reflectance. As objects present in a scene and lighting applications influence people’s judgement for colour rendition, the presented method is promising for colour rendering research.
The first section of the paper presents a database of hyperspectral images acquired using a VNIR4 SPECIM (sCMOS-50-V10E model) hyperspectral camera attached to a SPECIM rotating scanner. The database consists of a set of high spatial and spectral resolution images and is mainly composed of complex indoor scenes and semi-complex lab scenes. During the acquisition, special care was taken to be representative of multiple real life applications. In a second section, a new colour graphic icon informing on the colour rendering properties of light sources in context is presented. Instead of making predictions based on a predefined test colour samples, as traditionally done to characterise colour rendition of light sources, the specific colour content of the scene is analysed thanks to hyperspectral images. The proposed colour graphic icon is then applied to the scene. It provides accurate and reliable prediction of the effects of light sources on the rendering of the scene. The third section of the paper illustrates the interest of the graphic icon through various case studies. The results show the usefulness of such a tool to evaluate the colour rendering properties of different sources based on scene’s colour content.