Specialized orientation representation of neurons in three-dimensional spatiotemporal frequency domain

Neurons in the primary visual cortex (V1) are best known for their selectivity to orientation. Is orientation the most sensitive dimension among the stimulus parameters that influence their responses? To address this question, we aimed to analyze the responses of neurons in the cat primary visual cortex using a modified reverse correlation technique to obtain spectral receptive fields in the three-dimensional (3-D) spatiotemporal frequency domain. Comparison of tuning bandwidths revealed that neurons in the primary visual cortex were more sharply tuned to orientation than to spatial or temporal frequency, indicating that orientation was the most sensitive dimension in this stimulus space. Analysis of natural scenes showed that fine salient features were more elongated along the orientation axis than coarse ones. The same scale-dependent asymmetry between the orientation and orthogonal directions was observed in the tuning properties of these neurons in the two-dimensional (2-D) spatial frequency domain, suggesting that they became specialized for orientation through adaptation to natural image statistics. Most cat striate neurons had spectral receptive fields separable between the 2-D spatial frequency plane and temporal frequency domain, allowing them to represent local motion energy in the 3-D frequency domain. When the responses were compared across contrasts, complex cells signaled optimal orientation robustly (i.e., with minimal fluctuation) in the spectral receptive fields. Extensive spatial pooling of feature detectors along the orientation axis appeared to explain this property at least partially. Such spatial pooling may underlie the reliable signaling of visual inputs in noisy contexts as well as in position-tolerant representation.NEW & NOTEWORTHY Orientation-based representation is a key feature shared by the early stages of both biological and machine visual systems. We demonstrate that neurons in the primary visual cortex exhibit sharper tuning to orientation than to other stimulus parameters in the spectral domain and that their scale-dependent tuning parameters appear to inherit statistical properties from natural images. Furthermore, in noisy visual environments, orientation is signaled more reliably by complex cells that pool feature detectors extensively across space.