Detection of bilateral symmetry using spatial filters

Area of Science:

• Computational vision
• Neuroscience
• Image processing

Background:

• Symmetry detection is crucial for visual perception.
• Previous models focused on blob alignment or correlation for symmetry.
• Human symmetry detection mechanisms remain incompletely understood.

Purpose of the Study:

• To investigate the computational mechanisms underlying human symmetry detection.
• To compare different computational models of symmetry detection against human performance.
• To identify the key visual cues involved in perceiving symmetry.

Main Methods:

• Simulated four computational models based on filtering and symmetry measures (blob alignment, correlation).
• Tested models using variations in spatial jitter, dot matching, and symmetrical region properties.
• Compared model performance against human performance data across various conditions.

Main Results:

• The model using oriented filters and blob alignment best predicted human symmetry detection.
• Other models, including those using isotropic filters or correlation, showed discrepancies with human performance.
• Feature co-alignment in oriented filter outputs proved critical.

Conclusions:

• Human visual symmetry detection relies on the co-alignment of features processed by oriented filters.
• This mechanism may be part of a broader visual grouping system for tasks like object recognition.
• Symmetry perception can be viewed as an emergent property of general feature detection and grouping.