Structures of Physical and Visual Light Fields: Measurement, Comparison and Visualization

It is impossible to see the light in an empty space, we can only observe light as emission from a light source or reflections from objects. Yet, human observers can estimate the illumination in empty parts of an observed scene, based on the appearance of surrounding objects. This dissertation presents studies on human sensitivity to the light field structure in empty spaces and description of the development of a light visualization tool that implements our knowledge of light fields, light design and perception.

In our perceptual studies, we reconstructed and compared physical and visual light fields. Physical measurements of the illuminance were made in real and modelled scenes with Cuttle's cubic measurement approach. The measurement device was a cube (a simulated one for modelled scenes) with small sensors on each side. The device was positioned over a grid of points in scenes creating regular measurements. For each position six measurements were translated to light properties (intensity, vector direction, diffuseness) with Cuttle's formulas. Then the resulting data was interpolated in order to obtain a full light field. In psychophysical experiments we used a probe proposed by Koenderink et al., a white matte sphere on which the illumination could be controlled by an observer. The task was to make the probe visually fit to a scene or an object. Placing the probe over grids of positions we obtained user data that was proven to be robust enough to reconstruct the global visual light field. We demonstrated that observers' data is robust enough to reconstruct the global structure of the visual light field. We also found that the visual light field is simplified with respect to the physical truth. In particular, it does not reflect subtle variations of the physical light field. In studies on scenes with complex light field structures (i.e. light zones, neighboring light fields with contrasting differences in one or more light properties), we found that observers are quite sensitive to the difference in light properties between the light zones. However, they showed idiosyncratic behavior especially for light zones with diferences in depth of a scene (front-back), rather than in the picture plane (left-right).

The second goal of this thesis was to develop a tool that incorporates our knowledge in measurement and perception of light in its visualization. Modern light visualizations often focus on surfaces or show light in a sophisticated manner understandable only for experts. We augment existing approaches with our tool that visualizes light in 3D volumes and in a perceptually-relevant manner. The measurement approach was the same as the physical measurements used for the perceptual studies above. The measurement cubes could be implemented physically, for real, or virtually, for modelled scenes. Resulting measurements were translated into light properties - mean illuminance, vector direction and diffuseness of light - and represented via variation of shapes' proportions. We tested our visualizations performance compared to image renderings and found that the visualizations led to at least as good task performance as renderings. Moreover, we developed a web-based tool, which can be used for visualizing of cubic measurements by anyone and described applications of this tool for architectural lighting design.

Our findings expand knowledge on the structure of the visual light field and help to understand it better. This can contribute to applied areas, such as computer graphics and architectural lighting design. Moreover, our visualization tool can immediately be used by lighting artists or architectural light designers for increasing their work efficiency by providing quick and quantitative representation of the light conditions.