Neuromorphic visual systems which take cues from neural structure have already demonstrated impressive efficiency. This inspires people to examine the structure of the primary visual cortex (V1), which is the center of brain's visual signal processing and the self-organization mechanism that enables brain's amazing functional capabilities. The most significant property in V1 is orientation selectivity. However, its underlying neural mechanisms have yet to be fully illuminated and the origin of orientation selectivity in the responses of simple cells in the mammalian visual cortex serves as a model problem for understanding cortical circuitry and computation....[ Read more ]

Neuromorphic visual systems which take cues from neural structure have already demonstrated impressive efficiency. This inspires people to examine the structure of the primary visual cortex (V1), which is the center of brain's visual signal processing and the self-organization mechanism that enables brain's amazing functional capabilities. The most significant property in V1 is orientation selectivity. However, its underlying neural mechanisms have yet to be fully illuminated and the origin of orientation selectivity in the responses of simple cells in the mammalian visual cortex serves as a model problem for understanding cortical circuitry and computation.

We propose a developmental model of V1 orientation selectivity, where orientation selectivity develops through Hebbian rules applied to the geniculo-cortical connections. Cortical neurons are grouped into orientation columns and then coupled excitatorily by directly neighboring connections and inhibitorily through interneurons. Cortical coupling strengths are modeled by a Gaussian function and a one-dimensional cortical map structure emerges from the balanced interactions between excitation and inhibition, where inhibition enforces the coverage uniformity of the orientation distribution and excitation implements continuity. The resultant map demonstrates periodic orientation progression along the cortex and maintains the simple cell receptive field structures at the same time. This model establishes a link between biologically plausible Hebbian models and more abstract dimension reduction models for orientation preference. It also extends previous work to include contrast invariance. We demonstrate that incorporating contrast invariance improves the robustness of receptive field formation in the presence of mismatch and improves the continuity of the resultant orientation map.

Keywords- Hebbian learning, contrast-invariant tuning, orientation hypercolumns, cortical map formation, self organized map formation, dimension reduction, feedback inhibition