Related Experiment Videos
Quantification of local symmetry: application to texture discrimination
Y Bonneh1, D Reisfeld, Y Yeshurun
1Department of Computer Science, Tel Aviv University, Israel.
Spatial Vision
|January 1, 1994
Summary
This study shows that a computational model of generalized symmetry can predict human performance in texture discrimination tasks. The generalized symmetry transform effectively captures local spatial relationships in images, highlighting key features for visual perception and computer vision applications.
Area of Science:
- Computer Vision
- Computational Neuroscience
- Visual Perception
Background:
- Symmetry is a fundamental cue in visual perception and computer vision.
- Generalized Symmetry Transform (GST) identifies local symmetry intensity and orientation from edge maps.
- GST aids in detecting regions of interest for active vision systems.
Purpose of the Study:
- To evaluate the efficacy of the Generalized Symmetry Transform in predicting human performance in texture discrimination tasks.
- To demonstrate the role of local spatial relations in visual processing.
Main Methods:
- Applied the Generalized Symmetry Transform to edge maps of micro-patterns used in psychophysical experiments.
- Correlated the output symmetry map with human performance data in texture discrimination.
Main Results:
- The computational scheme based on generalized symmetry successfully accounted for most of the observed human performance in texture discrimination.
- The transform's ability to capture local spatial relations between image edges is key to its predictive power.
Conclusions:
- Generalized symmetry, as computationally defined, is a significant factor in human texture discrimination.
- The Generalized Symmetry Transform provides a valuable tool for understanding visual perception and developing computer vision algorithms.