Frequency tuning of shape perception revealed by classification image analysis

Area of Science:

• * Visual Perception and Cognitive Neuroscience
• * Machine Learning and Image Analysis

Background:

• * Classification image analysis is established for contrast detection but limited for shape discrimination.
• * Previous shape perception studies using this method were restricted to simple radial shapes.

Purpose of the Study:

• * To develop a novel classification image methodology for identifying linear mechanisms in general 2-D shape discrimination.
• * To apply this method to arbitrary and natural shapes, extending beyond simple radial forms.

Main Methods:

• * Projecting target shapes onto a Fourier descriptor (FD) basis set to represent key perceptual features.
• * Utilizing a yes/no paradigm to efficiently identify the observer's classification template.
• * Matching stimulus noise spectral density in FD space to the target shape's power law density.

Main Results:

• * Demonstrated that natural shapes exhibit low-pass characteristics under FD projection, following a power law.
• * Developed linear template models for animal shape detection that predict human judgments.
• * Observed that classification templates are biased, overweighting lower frequencies.

Conclusions:

• * The proposed method successfully models linear mechanisms for complex shape discrimination.
• * The identified low-pass bias suggests that higher-frequency shape information processing involves nonlinear mechanisms.