Segmentation of object outlines into parts: a large-scale integrative study

Area of Science:

• Computer Vision
• Cognitive Science
• Psychophysics

Background:

• Object segmentation is crucial for visual perception.
• Previous research proposed various rules for part formation, but empirical validation was limited.

Purpose of the Study:

• To empirically validate existing object segmentation rules.
• To develop an integrated framework for part formation based on human observer data.

Main Methods:

• 201 human observers segmented 88 line drawings of everyday objects.
• Collected data served as a benchmark for evaluating segmentation models.
• Analyzed observer strategies, focusing on contour features and global shape properties.

Main Results:

• Human segmentation frequently utilized minima of curvature along object contours.
• Global shape characteristics like proximity, collinearity, symmetry, and elongation influenced segmentation.
• Previously proposed segmentation rules were empirically supported by observer data.

Conclusions:

• A unified framework integrating validated segmentation rules is proposed.
• This framework enhances understanding of human object part formation.
• Provides a benchmark for developing more robust computer vision segmentation algorithms.