Shape from shading. I: Surface curvature and orientation

Area of Science:

• Visual perception
• Computational vision
• Psychophysics

Background:

• Shading is a crucial monocular cue for inferring 3D object shape.
• Its effectiveness for depth perception is debated compared to binocular disparity and motion cues.
• Understanding human shape-from-shading mechanisms informs computer vision algorithms.

Purpose of the Study:

• To quantify human curvature discrimination precision using shading.
• To investigate the influence of light source position and texture on shape perception.
• To evaluate computational models against human performance in shape-from-shading.

Main Methods:

• Measured curvature discrimination thresholds using a surface-alignment task.
• Varied surface curvature, light source rotation, and texture contrast.
• Compared psychophysical data with computer vision algorithms for shape-from-shading.

Main Results:

• High precision (Weber fractions ~0.1) in curvature discrimination from shading.
• Light source rotation towards the viewer and added texture degraded performance.
• Texture acted as noise, increasing thresholds even when consistent with shading cues.

Conclusions:

• Human visual system robustly extracts shape from shading, with curvature as a primary representation.
• Texture can interfere with shape-from-shading, acting as a noise source.
• Koenderink's local-shape index proposed as a model for human intermediate shape representation.