Lynn Schmittwilken¹, Anna, L. Haverkamp¹, Marianne Maertens¹; ¹Technische Universität Berlin

Edges are 2d image features that are indicative of object boundaries in the natural world. To understand how humans extract edges from the visual input, many psychophysical studies characterize edge sensitivity in well-controlled tasks and with parametrized stimuli. While this approach allows to test specific hypotheses about the underlying mechanisms, it is not clear how well these insights translate to contour perception in natural scenes. Here we test how contour perception in natural images relates to edge sensitivity when both types of stimuli are perturbed with different types of noise. We compare edge sensitivity data from a 2-AFC task with simple stimuli, and line drawings from a contour tracing task in images. In both cases, we perturb the stimuli with 2d noise to probe the putative underlying spatial frequency (SF) selective mechanisms. We used three broadband noises (white, pink, brown) and three narrowband noises (center SFs: 0.5, 3, 9 cpd). In the 2-AFC task, observers indicated the location of a Cornsweet edge via a button press. We used three Cornsweet edges with different peak SFs (0.5, 3, 9 cpd). In the contour tracing task, subjects drew all visible contours in a variety of natural scenes from the Contour Image Database (Grigorescu et al., 2003) using a drawing tablet. In both cases, we measured psychometric functions varying edge contrast. In the drawing task, we quantified performance as the amount of agreement between the contour traces in the presence and absence of noise. Finally, we compared the corresponding noise masking effects in both tasks. Our results provide an example that early visual processes can be investigated in more behaviorally-relevant and meaningful tasks. Furthermore, our data serves as a useful augmentation of the Contour Image Database to study the mechanisms underlying human edge sensitivity via computational modeling.