Local figure-ground cues are valid for natural images

Area of Science:

• Visual Perception
• Computational Neuroscience
• Image Analysis

Background:

• Figure-ground organization is a fundamental aspect of visual perception, determining how contours are assigned to regions.
• Recent research indicates early neural processing (V2, 10-25 ms post-response) supports bottom-up figure-ground assignment.

Purpose of the Study:

• To investigate the information content of local cues for figure-ground assignment.
• To quantify the ecological validity and relative importance of Gestalt cues (size, convexity, position).
• To develop and evaluate a computational model for figure-ground assignment based on local image statistics.

Main Methods:

• Analysis of a large dataset of natural images with human-assigned figure-ground relations.
• Quantification of local image properties such as region size, convexity, and relative position.
• Development of a bottom-up computational model using image contours and fitted parameters.

Main Results:

• Figural regions are statistically more likely to be smaller, more convex, and located below ground regions in natural images.
• These Gestalt cues are ecologically valid and their relative contributions were quantified.
• The computational model demonstrated human-level performance in figure-ground assignment.

Conclusions:

• Local bottom-up cues, particularly Gestalt principles, provide substantial information for figure-ground organization.
• A computational model based on these local cues can effectively replicate human visual perception.
• The findings underscore the significance of local processing in early visual perception.