Fixational eye movements enable robust edge detection.

Area of Science:

• Computational Neuroscience
• Computer Vision
• Human Vision Research

Background:

• Human vision processes luminance edges using spatial mechanisms across multiple scales.
• Existing models typically assume static inputs and ignore temporal modulations from fixational eye movements.
• Recent findings highlight the functional importance of temporal input modulations caused by eye movements.

Purpose of the Study:

• To propose a spatiotemporal model of human edge detection integrating spatial and active vision principles.
• To investigate the role of temporal filtering and active sampling via fixational eye movements in edge detection.
• To challenge the necessity of orientation-selective mechanisms in edge detection.

Main Methods:

• Developed a spatiotemporal model augmenting a spatial vision model with temporal filtering and image shifting.
• Tested the model on White's illusion by superimposing narrowband noise.
• Evaluated edge detection performance in natural images with and without Gaussian white noise, comparing with human-labeled contours.

Main Results:

• The model accurately reproduced the spatial-frequency-specific interference observed in White's illusion.
• Demonstrated robust edge detection against noise in natural images, independent of orientation-selective processes.
• Confirmed the importance of multiscale spatiotemporal filtering and scale-specific normalization through component elimination.

Conclusions:

• The proposed spatiotemporal model offers efficient edge detection for artificial vision systems.
• Fixational eye movements and spatiotemporal filtering are crucial for robust edge detection.
• Orientation-selective mechanisms are not strictly required for effective edge detection.