Visual perception of procedural textures: identifying perceptual dimensions and predicting generation models

Area of Science:

• Computer Graphics
• Human-Computer Interaction
• Psychophysics

Background:

• Procedural models generate realistic textures but lack perceptual meaning for users.
• Artists and designers prefer intuitive descriptors like 'repetitive' or 'directional' for texture creation.

Purpose of the Study:

• Investigate the perceptual dimensions of procedurally generated textures.
• Develop a perceptually meaningful framework for texture synthesis.
• Bridge the gap between mathematical models and user intuition in texture design.

Main Methods:

• Conducted two psychophysical experiments: free-grouping and rating.
• Applied Hierarchical Cluster Analysis (HCA) and Singular Value Decomposition (SVD) to analyze texture groupings.
• Utilized isometric feature mapping (Isomap) to construct a 3D perceptual texture space.

Main Results:

• Identified limitations of existing perceptual dimensions for random textures.
• Established a robust 3D perceptual texture space explaining human similarity judgments.
• Developed computational models linking perceptual features to the new texture space.

Conclusions:

• The proposed 3D perceptual texture space effectively captures human texture perception.
• Computational models allow procedural texture generation based on user-defined perceptual scales.
• This work enhances usability of procedural texture models for artists and designers.