Shape-from-shading and light source estimation in humans

Mazzilli, Giacomo (2015). Shape-from-shading and light source estimation in humans. University of Birmingham. Ph.D.

Preview

Mazzilli15PhD.pdf
PDF
Download (16MB)

Abstract

Light source estimation is very important for the interpretation of shape-fromshading by humans. We used a range of methods to characterise the way in which the type, and position of the light source can influence observers’ performance in shape-from-shading tasks.
Firstly, we used classification images to discover people’s priors for light source position using noise only stimuli. This cue-free approach uncovered the weakness of the light-from-above prior. We also examined the effect of varying the light source elevation on the perceived shape of isotropic and anisotropic surfaces, the impacts of lighting ambiguities on shape-from-shading and, finally, the interpretation of shadow regions.
We found that lighting priors are weighted by the visual system in a way that is inversely proportional to the strength of lighting cues in the stimuli, revealing that knowledge about the light source position is critical to perceiving shapefrom- shading. Where ambiguous cues to lighting direction are present human vision seems to favour local cues over distal ones. We also showed that perceived surface shape varies with light elevation only in so far as elevation alters contrast. Finally we show that human vision does not treat shadows in the same way as objects.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):