Is edge sensitivity more than contrast sensitivity?

How does the human visual system extract relevant information from the pattern of light on the retinae? The psychophysical study of early vision has led to important insights in the initial processing characteristics of human vision, and we have image-computable models which quantitatively account for relevant psychophysical findings. Many of these findings come from experiments with sinusoidal gratings and Gabor patches, which, due to their narrow spectra, are thought to be ideal to isolate and test mechanisms of early vision. However, during natural viewing relevant stimulus features such as object boundaries involve sharp luminance discontinuities (i.e. edges) and thus broad spectra. Here we explore whether the computational mechanisms which account for the perception of sinusoidal gratings can also predict psychophysical sensitivity to (isolated) edges. Psychophysically, we probe human edge sensitivity (2-AFC, position of edge) in the presence of different types of noise. Edges were Cornsweet edges with peak frequencies at 0.5, 3 and 9 cpd, and noise types were white, pink, and brown as well as three narrowband noises with center frequencies at 0.5, 3 and 9 cpd. Computationally, we implemented several edge models using standard components of existing edge models (single or multiple odd-symmetric log-Gabor filters) and standard components from spatial vision models (divisive contrast gain control, signal-detection-theory based decoders of varying complexity). We find that several different models can reasonably account for the data and thus conclude that human contrast and edge sensitivity are likely employing similar computational mechanisms - results from narrowband sinusoidal gratings thus generalize to more natural, broadband stimuli. Finally, we discuss the assumptions and design choices involved in our computational modeling and make both explicit to foster their critical examination.