Neural computation of visual imaging based on Kronecker product in the primary visual cortex

Area of Science:

• Neuroscience
• Computational Neuroscience
• Image Processing

Background:

• The precise neural computation within the primary visual cortex (V1) and its mathematical representation remain key unanswered questions in visual information processing.
• Understanding V1's function is crucial for deciphering how the brain processes visual input.

Purpose of the Study:

• To mathematically model the neural computation performed by the primary visual cortex (V1) for visual image reconstruction.
• To propose a novel algorithm based on retinal organization and topographical mapping for V1 image processing.

Main Methods:

• Dividing images into orthogonal arrays of image primitives (patches) that activate simple cells in V1.
• Utilizing the inner product, specifically the Kronecker product, for neural computation between image patches and V1's functional architecture.
• Employing computer simulations with two-dimensional Gabor pyramid wavelets to validate the proposed model.

Main Results:

• The neural computation in V1 can be effectively represented by Kronecker product operations and their matrix forms.
• The Kronecker product algorithm is simple, efficient, robust, and suitable for biological implementation in the visual cortex.
• Computer simulations confirmed the validity of the theoretical analysis and the proposed model.

Conclusions:

• The Kronecker product offers a computationally efficient and biologically plausible mechanism for neural computation in V1, replacing complex Fourier Transforms.
• Cortical columns in V1 are better understood as basic units for processing visual primitives (e.g., contours, edges) rather than simple filter arrays.
• This model provides significant insights into the neural mechanisms underlying human visual information processing.