Keywords

Shape, efficient coding, curvature, V4, neural coding

Abstract

Efficient coding provides a concise account of key early visual properties, but can it explain higher-level visual function such as shape perception? If curvature is a key primitive of local shape representation, efficient shape coding predicts that sensitivity of visual neurons should be determined by naturally-occurring curvature statistics, which follow a scale-invariant power-law distribution. To assess visual sensitivity to these power-law statistics, we developed a novel family of synthetic maximum-entropy shape stimuli that progressively match the local curvature statistics of natural shapes, but lack global structure. We find that humans can reliably identify natural shapes based on 4th and higher-order moments of the curvature distribution, demonstrating fine sensitivity to these naturally-occurring statistics. What is the physiological basis for this sensitivity? Many V4 neurons are selective for curvature and analysis of population response suggests that neural population sensitivity is optimized to maximize information rate for natural shapes. Further, we find that average neural response in the foveal confluence of early visual cortex increases as object curvature converges to the naturally-occurring distribution, reflecting an increased upper bound on information rate. Reducing the variance of the curvature distribution of synthetic shapes to match the variance of the naturally-occurring distribution impairs the linear decoding of individual shapes, presumably due to the reduction in stimulus entropy. However, matching higher-order moments improves decoding performance, despite further reducing stimulus entropy. Collectively, these results suggest that efficient coding can account for many aspects of curvature perception.

Share

COinS
 

Efficient Coding of Local 2D Shape

Efficient coding provides a concise account of key early visual properties, but can it explain higher-level visual function such as shape perception? If curvature is a key primitive of local shape representation, efficient shape coding predicts that sensitivity of visual neurons should be determined by naturally-occurring curvature statistics, which follow a scale-invariant power-law distribution. To assess visual sensitivity to these power-law statistics, we developed a novel family of synthetic maximum-entropy shape stimuli that progressively match the local curvature statistics of natural shapes, but lack global structure. We find that humans can reliably identify natural shapes based on 4th and higher-order moments of the curvature distribution, demonstrating fine sensitivity to these naturally-occurring statistics. What is the physiological basis for this sensitivity? Many V4 neurons are selective for curvature and analysis of population response suggests that neural population sensitivity is optimized to maximize information rate for natural shapes. Further, we find that average neural response in the foveal confluence of early visual cortex increases as object curvature converges to the naturally-occurring distribution, reflecting an increased upper bound on information rate. Reducing the variance of the curvature distribution of synthetic shapes to match the variance of the naturally-occurring distribution impairs the linear decoding of individual shapes, presumably due to the reduction in stimulus entropy. However, matching higher-order moments improves decoding performance, despite further reducing stimulus entropy. Collectively, these results suggest that efficient coding can account for many aspects of curvature perception.