Frequency-based heuristics for material perception.

Area of Science:

• Visual perception
• Computational neuroscience
• Material science

Background:

• Humans make quick visual judgments about surface properties for interaction.
• Complex image formation challenges inverse optics algorithms without external constraints.
• Heuristics based on retinal image information may be employed.

Purpose of the Study:

• To investigate how visual systems perceive 3-D material properties from 2-D images.
• To identify the role of luminance variations and spatial frequencies in material perception.
• To explore the implications for machine vision and computer graphics.

Main Methods:

• Perceptual scaling of a large image dataset.
• Correspondence and canonical correlation analyses.
• Psychophysical experiments using movies and adaptation to dynamic noise.

Main Results:

• Material properties (roughness, thickness, undulations) correlate with specific luminance variation scales.
• Perception of 3-D qualities changes with relative energy in 2-D frequency bands.
• Adaptation to noise at specific scales predictably alters material judgments.

Conclusions:

• The scale of local 3-D structure is critical for perceiving material properties.
• Relative contrast at specific spatial frequencies is key for 3-D structure-from-shading perception.
• Cortical mechanisms may combine frequency-selective neuronal outputs for material estimation.