Abstract
Crowding refers to the inability to recognize an object surrounded by other objects. One account of the phenomenon is the inappropriate integration of object features within an integration field. However, little is known about how features from a target and its flankers are being integrated. In this study, we derived a set of combination rules of features based on human empirical data. We measured observers’ accuracy for identifying a target symbol flanked by two other symbols (4000 trials per observer), presented for 100 ms at 10° below fixation. Our symbols (a set of 10) were constructed with one or two "features" — single line segments (0.5° in length) oriented at 0°, 45°, 90° or 135°. Averaged across three observers, performance accuracy for identifying single symbols was 84%, and dropped to 50% in the presence of flankers, demonstrating a crowding effect. Analysis of the error trials revealed that (1) the probability of a given segment being present in the response increased with its frequency of occurrence in the flankers; (2) error rate increased with flanker complexity (total number of flanker features); and (3) error rate increased with the number of shared features (similarity) between the target and its flankers. A model based on the human identification performance for unflanked symbols and a weighting term to represent segment positional uncertainty (72% weight for the target and 14% weight for each of the flankers) well predicts the segment frequency and the complexity effects (chi-squared test, p<0.001), contrary to a model based on the positional uncertainty for the whole symbol (chi-squared test, p>0.05). However, to capture the similarity effect, the relative weighting between the target and the flankers must decrease with increased target-flanker similarity. These results suggest that positional uncertainty of target and flanker features could account for how features are integrated erroneously during crowding.
Meeting abstract presented at VSS 2012