Perception of three-dimensional shape from texture is based on patterns of oriented energy

Area of Science:

• Computer Vision
• Perception Psychology
• Computational Geometry

Background:

• Inferring 3D shape from 2D images is a fundamental problem in computer vision and human perception.
• Monocular texture cues offer a rich source of information for shape perception.

Purpose of the Study:

• To present empirical support for a novel observer model of 3D shape inference from monocular texture cues.
• To investigate the role of local spectral changes and oriented energy in texture patterns for conveying 3D surface curvature.

Main Methods:

• Observers estimated 3D curvature from 2D projected images of textured surfaces.
• Local spectral changes in projected images were analyzed for various texture patterns (gratings, plaids, noise).
• Perceived ordinal depth was reconstructed from measurements of local depth at different phases of a corrugated surface.

Main Results:

• Correlated changes in oriented energy along lines of maximum and minimum curvature are crucial for 3D shape perception.
• Identifying individual texture elements or gradients is neither necessary nor sufficient for shape extraction.
• Veridical depth perception requires changes in oriented energy along projected lines of maximum curvature.
• Perspective projection is essential; parallel projection loses critical texture cues for shape inference.

Conclusions:

• Texture is a valuable cue for 3D shape inference in perspective projections.
• The global Fourier transform of natural textures can predict their suitability for veridical shape inference.
• Future models should focus on oriented energy changes along curvature lines for robust 3D shape reconstruction from texture.