Medium spatial frequencies mask edges most effectively

Luminance discontinuities, i.e. edges, are important visual image features because they demarcate the boundaries of objects in the physical world. Psychophysically, human edge sensitivity has been studied in different ways including edge localization, edge polarity judgments, near-threshold edge detection or reverse correlation techniques. Different methods have yielded different results in particular with respect to the question to which extent edge sensitivity can be accounted for by a multi- or a single-scale edge detector model. Here we test whether edge sensitivity is mediated equally by a wide range of spatial frequency selective channels or whether it is mainly mediated by a more narrow spatial scale. We probe edge sensitivity in the presence of noise that specifically interferes with one or several of the presumed spatial scales as this is an effective way to constrain respective models. We test psychophysically how well observers can detect a step edge in the presence of noise with varying spectral properties: white noise, pink noise (1/f), brown noise (1/f-squared) and three types of narrowband noise with peak spatial frequencies of either 0.58 cpd, 3 cpd or 9 cpd. Our results show two main effects: (1) Noise with large spectral power close to the peak spatial frequency of the edge stimulus effectively reduced human edge sensitivity. (2) Noise with large spectral power around medium spatial frequencies of 3 cpd effectively reduced human edge sensitivity independent of the peak spatial frequency of the edge stimulus. Our results support the notion that a relatively narrow spatial scale around 3 cpd plays a special role in extracting edge information from the visual input.