Multidimensional gain control in image representation and processing in vision

Area of Science:

• Neuroscience
• Computer Vision
• Sensory Perception

Background:

• Automatic gain control (AGC) is crucial for adjusting sensory sensitivity across wide dynamic ranges.
• Existing models often focus on single modalities or specific functions.

Purpose of the Study:

• To propose a generic, multidimensional AGC model for contrast sensitivity adjustment in vision and other senses.
• To investigate the role of AGC in enhancing image attributes and explaining visual illusions.

Main Methods:

• Developing a generic feedback AGC mechanism with nonlinear synaptic interaction in a neural network.
• Simulating AGC to reproduce psychophysical data on size and depth contrast effects.
• Analyzing curvature processing by the AGC network for image structure pre-emphasis.

Main Results:

• The AGC model enhances image attributes like curvature, size, and depth, improving contrast.
• Simulations successfully replicated psychophysical data for the Ebbinghaus illusion and depth contrast effects.
• The model demonstrates AGC's role in image structure pre-emphasis and saliency enhancement.

Conclusions:

• A unified, parsimonious AGC neural network model explains neurobiological contrast control.
• This model accounts for visual illusions, contour completion, and other psychophysical phenomena.
• Biologically motivated AGC offers novel approaches for intelligent computer vision systems.