A Computational Model that Realizes a Sparse Representation of the Primary Visual Cortex V1

On the basis of synchronous oscillation in the visual cortex and synchronized responses to external stimuli, we have proposed a complete neural computational model of visual information processing, which consists of multiscale filtering, phase synchronization, and inner-product formation. In the model, firing-spike trains are topologically mapped from the retina to the cortex V1 and are synchronously decoded by neural phase-locked loops (NPLLs), and then the model forms an inner product of the outputs of the NPLLs with the receptive fields of simple cells, which are densely distributed in the visual cortex. The inner-product operation leads these simple cells to fire; the simple cells in a firing state form an activation pattern that is a reconstruction of the image of the external visual stimulus. This computational model reveals clearly a computational process of inner-product formation that is an effective approach to realizing a sparse representation. The multiscale filtering, decoding, and inner-product operations on the visual image reflect the main properties of visual information processing, such as efficiency, simplicity, and robustness from the point of view of neural computation. This finding provides a neural computation suitable for realizing a sparse representation of external visual images and provides further insight into information processing in V1.