HDR versus hyperspectral images for applied colour and lighting research

For a long time, images have been employed in the architectural process to judge the aesthetic qualities of designed spaces, and to communicate with clients. They are now also increasingly used by colour and lighting researchers for investigating people's perception of lighting environments and daylit spaces. Among their advantages, we can cite the control of experimental conditions and of investigated parameters, and also the reduction of cost in comparison to real space, mock-up or reduced-scale model. Previous works have shown that, to some extent, both photographs and computer generated-pictures have the potential to predict how the not yet built environment will be experienced. As long as perceptual attributes induced by the displayed pictures are faithful to reality, and the physical world is accurately captured by the camera or rendered by the software, images can be used as visual stimuli in a psychophysical approach. Nevertheless, each step in the production of pictures – from scene creation to image visualisation – can lead to inaccuracies. The present study proposes to assess the colorimetric accuracy of three types of images: - HDR photographs The main advantage of such images is that they can be produced with a simple digital camera. For more than ten years, high dynamic range (HDR) photography has increasingly been used as a luminance data acquisition tool by lighting researchers investigating quality of lighting and visual comfort. A deviation of luminance of less than 10% with peak values reaching up to 20% or sometimes higher can be expected from HDR photography. Few data are available in the literature regarding colour accuracy. - Hyperspectral photographs Hyperspectral camera manufacturers claim to produce high definition images while capturing for each spatial pixel spectral information with an optical resolution equivalent to spot spectroradiometers. Such a technology is interesting for lighting and colour research in architecture because from the acquired radiances, spectral reflectances can be calculated. The initial scene can then be simulated under different real or virtual illuminants/glazing. - Hyperspectral images computationally rendered In addition to the aforementioned advantages of hyperspectral imaging, simulation makes it possible to investigate not yet built spaces without requiring a real reference scene. Cost are thus further reduced and experimental conditions or investigated parameters can be better controlled. However, most software are complex and sometimes difficult to use. The current lack of material properties database can still lead to discrepancies between real and rendered materials. In the present study, photometric and colorimetric accuracy is quantified on a Macbeth chart lit by sources with various CCTs. The reference is spot measurement with a spectroradiometer. Colour differences are calculated in CIELAB. Image sharpness is also analysed