Neural Basis of Perceptual Surface Qualities and Materials: Evidence from Electroencephalogram Decoding

Area of Science:

• Neuroscience
• Computer Vision
• Psychophysics

Background:

• Human visual system excels at material and surface property recognition.
• Material perception is influenced by global feature statistics.
• Understanding neural dynamics of material perception is crucial.

Purpose of the Study:

• Investigate dynamic neural representations of material and surface properties.
• Classify material categories and surface properties from visual evoked potentials (VEPs).
• Correlate VEPs with global image features and reconstruct surfaces from neural data.

Main Methods:

• Measured VEPs from 191 natural images across 20 material categories.
• Classified material categories and surface properties (lightness, colorfulness, smoothness, glossiness, hardness, heaviness) from VEPs.
• Employed reverse-correlation analysis and deep generative models (multimodal variational autoencoders).

Main Results:

• Material categories classified from VEPs within 150 ms.
• Surface properties classified within 175 ms (lightness, colorfulness, smoothness) and after 200 ms (glossiness, hardness, heaviness).
• VEPs correlated with low- and high-level global image features; reconstructed images matched original material properties.

Conclusions:

• Early cortical responses reflect statistical features crucial for material perception.
• Neural representations of global features rapidly support material recognition.
• Deep generative models can reconstruct perceptually relevant surface information from neural data.