Statistics of lines of natural images and implications for visual detection

Abstract

As borders between different regions, lines are an important element of natural images. Already at the level of the mammalian primary visual cortex (V1), neurons respond best to lines of a given orientation. We reduce a set of images to linear segments and analyze their statistical properties. In particular, appropriately defined Fourier spectra show more power in their transverse component than in the longitudinal one. We then characterize filters that are best suited for extracting information from such images, and find some qualitative consistency with neural connections in V1. We also demonstrate that such filters are efficient in reconstructing missing lines in an image.