Colour appearance modelling of unrelated self-luminous stimuli

Colour appearance models, i.e. models that attempt to predict the colour appearance of a stimulus by taking the physical properties of the stimulus and its surroundings into account, have been developed and investigated for more than 40 years. Most of these models were developed to handle related colours, i.e. colours perceived in relation to other colours. A typical example is the ‘reflected’ colour of an object as seen in an illuminated scene. However, two models - CAM97u and CAMFu - were developed to predict the appearance of unrelated colours, i.e. colours perceived in isolation from any other colour (e.g. a traffic light seen at night). Unfortunately, due to the lack of psychophysical data, neither of these two models has been investigated extensively. Before being able to extend these models to other types of stimuli and viewing conditions, they need to be validated using new visual data.

The aim of this doctoral research is toinvestigate the colour appearance of unrelated self-luminous stimuli. An accurate prediction of the colour appearance of these stimuli through a colour appearance model can be a valuable tool: it can assist in the development of requirements for light-emitting diode (LED) signs, in the standardization of the appearance of marine, aviation or traffic lights viewed during a dark night, in the continuous development of colour appearance models for other viewing conditions,…

In a first series of psychophysical experiments, the brightness of stimuli with a constant luminance has been evaluated by a group of observers. The stimuli were shown in a darkened room, specially designed for this doctoral research project. In the centre of one wall, a circular self-luminous area was present. The colour of this stimulus area was computer controllable by adjusting the flux of the R(ed)G(reen)B(lue)W(white) LED behind it. The observers viewed the stimulus area from a distance that ensures a 10° field of view. The brightness evaluation of these stimuli was performed using a magnitude estimation method by scaling the brightness of each test stimulus compared to that of a reference stimulus to which a brightness value of 50 was attributed. The predictive performance of the CAM97u and CAMFu colour appearance models and four other vision models, specially designed to predict brightness, was investigated. Due to, among others, a severe underestimation of the effect of colourfulness on brightness - also known as the Helmholtz-Kohlrausch effect - none of the models seemed to be able to adequately predict the brightness perceived by the observers. Adapting the CAM97u model by increasing the colourfulness contribution in the brightness attribute, resulted in a modified model, called CAM97um, which allows for a substantially better brightness prediction. The performance of this new model was confirmed by the results of both a matching experiment and an extensive magnitude estimation experiment in which the test stimuli covered a wide range of luminance and chromaticity values.

In a subsequent series of psychophysical experiments, in addition to the brightness, the hue and “amount of white” perception of unrelated self-luminous stimuli was also investigated using a magnitude estimation method. The amount of white is a newly proposed attribute, and basically corresponds to a layperson’s conception of attributes such as colourfulness, chroma or saturation. It was introduced based on the results of a preliminary pilot study revealing that laypersons often have difficulty understanding, and hence judging, the colourfulness of a stimulus. Again, unrelated self-luminous 10° stimuli, with a wide range of luminance and chromaticity values, were evaluated by observers in the darkened room. Based on the obtained visual data, a new colour appearance model for unrelated self-luminous stimuli, CAM15u, was developed. The main features of the model are the use of the absolute spectral radiance of the stimulus as input, the use of the CIE 2006 cone fundamentals and a simplified calculation procedure compared to existing models. The model predicts the brightness, hue, colourfulness, saturation and the amount of white. The CAM15u model is restricted to photopic, non-glary unrelated stimuli having a field of view of 10°. The model was validated using the results of an additional experiment. It was found that, despite its simplicity, CAM15u performs as well or better than other, more complicated, CAMs.

In a final series of psychophysical experiments, the brightness perception of different sized, unrelated self-luminous stimuli was investigated in a magnitude estimation experiment. The stimuli were shown in a darkened room on a wide gamut LCD monitor. A significant, hue independent, effect of stimulus size on brightness was found, effectively modeled by a simple power function. Finally, the dependence of brightness on stimulus size was incorporated into the brightness prediction of the CAM15u model. The predictive performance of the modified brightness prediction was validated using the results obtained in an additional experiment in which observers evaluated the brightness of unrelated self-luminous test stimuli with variable size, chromaticity and luminance.

Although further improvements and extensions are still possible, CAM15u has proven its value in predicting the appearance of unrelated self-luminous stimuli. It can be a valuable tool for the improvement of existing standards and guidelines for traffic signs, LED billboards,…