Measuring the visual salience of alignments by their non-accidentalness

Abstract

Quantitative approaches are part of the understanding of contour integration and the Gestalt law of good continuation. The present study introduces a new quantitative approach based on the a contrario theory, which formalizes the non-accidentalness principle for good continuation. This model yields an ideal observer algorithm, able to detect non-accidental alignments in Gabor patterns. More precisely, this parameterless algorithm associates with each candidate percept a measure, the Number of False Alarms (NFA), quantifying its degree of masking. To evaluate the approach, we compared this ideal observer with the human attentive performance on three experiments of straight contours detection in arrays of Gabor patches. The experiments showed a strong correlation between the detectability of the target stimuli and their degree of non-accidentalness, as measured by our model. What is more, the algorithm's detection curves were very similar to the ones of human subjects. This fact seems to validate our proposed measurement method as a convenient way to predict the visibility of alignments. This framework could be generalized to other Gestalts.

Generation and Detection of Alignments in Gabor Patterns

“…This parameterless algorithm is based on the a contrario theory, a mathematical model of the non-accidentalness principle. These methods were applied and tested in psychophysical experiments [2], where the NFA showed strong correlation with the detectability for human subjects, and the algorithm proved to be an accurate predictor of the average subject's behavior. Therefore, the framework presented here is a tool to test the relevance of the a contrario approach to perceptual tasks in image processing and computer vision.…”

“…This is what was done for the stimuli described in [2]. In the latter method, to produce one stimulus, ten were generated and the one with the greatest p-value was kept.…”

“…When the a contrario theory is applied to computer vision, the common practice is to detect only structures with NFA below a certain threshold (typically, NFA = 1). Like in [2], we want to compare the detection algorithm to visual perception on an objective task, in which both the subject and the algorithm are forced to report the most salient structure in the stimulus. If we allowed the algorithm to report no detection at all, we should give the same possibility to human observers, and this would introduce more subjectivity: the subject could actually see something and decide not to consider it as a significant alignment.…”

“…In other words, we have not found the appropriate trade-off between curvature and consistency within orientations. The algorithm presented here is, among the approaches we tried, the one that best matches the human data [2].…”

“…However, in general, the algorithm provides quite accurate predictions of the perception of alignments in our stimuli. For a detailed presentation of these results, we kindly refer to [2]. .…”