Line drawings as a tool to probe edge sensitivity in natural scenes.

To interact with the world effectively, the human visual system must extract meaningful features from visual scenes. One key feature are edges, luminance or texture discontinuities in two-dimensional (2D) images that often correspond to object boundaries in three-dimensional scenes. Edge sensitivity has traditionally been studied with well-controlled stimuli and binary choice tasks, but it is unclear how well these insights transfer to real-world behavior. Recent studies have extended this approach using natural images but typically retained binary button presses. In this study, we extend the approach further and ask observers (N = 20) to trace edges in natural scenes, presented with or without 2D visual noise. To quantify edge detection performance, we use a signal detection theory-inspired approach. Participants' edge traces in the noise-free condition serve as an individualized "ground-truth" or signal, used to categorize edge traces from noise conditions into hits, false alarms, misses, and correct rejections. Observers produce remarkably consistent edge traces across conditions. Noise interference patterns mirror results from traditional edge sensitivity studies, especially for edges with spectral properties similar to natural scenes. This suggests that insights from controlled paradigms can transfer to naturalistic ones. We also examined edge traces to identify which image features drive edge perception, using interindividual variability as a pointer to relevant features. We conclude that line drawings are a powerful tool to investigate edge sensitivity and potentially other aspects of visual perception, enabling nuanced exploration of real-world visual behavior with few experimental trials.