Multi-Frequency Image Completion via a Biologically-Inspired Sub-Riemannian Model with Frequency and Phase

Area of Science:

• Computational neuroscience
• Computer vision
• Image processing

Background:

• Image completion is crucial for restoring corrupted or missing image data.
• Existing methods often struggle to capture complex image features like orientation and frequency.
• The visual cortex processes visual information using specialized cells sensitive to these features.

Purpose of the Study:

• To develop a novel image completion algorithm inspired by the human visual cortex.
• To model the orientation, frequency, and phase-selective behavior of cortical cells.
• To leverage sub-Riemannian cortical geometry for enhanced image reconstruction.

Main Methods:

• Utilized a five-dimensional sub-Riemannian cortical geometry model.
• Extracted image features (orientation, frequency, phase) using a Gabor transform.
• Represented feature information as cortical cell output responses.
• Performed image completion via diffusion along neural connections in the model geometry.
• Transformed completed responses back to the image plane.

Main Results:

• The algorithm successfully reconstructed corrupted image regions.
• The cortically-inspired model effectively integrated orientation, frequency, and phase information.
• Diffusion process simulated neural activity propagation for feature integration.

Conclusions:

• The novel algorithm demonstrates a promising approach to image completion.
• Cortical geometry provides a robust framework for modeling visual information processing.
• This method offers a biologically plausible mechanism for image reconstruction.