Human Texture Vision as Multi-Order Spectral Analysis.

Area of Science:

• Neuroscience
• Computer Vision
• Perception

Background:

• Visual texture perception is crucial for recognizing scenes, objects, and materials.
• Existing models, like the Filter-Rectify-Filter (FRF) and Portilla-Simoncelli (PS) statistics, offer frameworks for understanding texture representation.
• Hierarchical processing in the visual cortex suggests multi-order spectral analysis is a viable approach.

Purpose of the Study:

• To propose a novel model for visual texture perception.
• To identify the critical spectral components determining texture perception.
• To provide a simpler framework for predicting texture representations in the primate visual system.

Main Methods:

• Extending the dimensionality of the Filter-Rectify-Filter (FRF) model.
• Analyzing 1st-order (2D-luminance) and 2nd-order (4D-energy) spectra.
• Preserving these key spectra while randomizing image phases to test perceptual similarity.

Main Results:

• Visual texture perception is primarily determined by the 1st-order luminance and 2nd-order energy spectra.
• Preserving these two spectra results in perceptually similar textures, validating the model.
• The proposed model aligns with the frequency representation of Portilla-Simoncelli (PS) statistics.

Conclusions:

• A simplified model based on two spectral spaces effectively describes primate visual texture representation.
• This spectral analysis approach is consistent with the hierarchical processing principles of the visual cortex.
• The model provides a foundation for further research into visual perception and artificial neural networks.