Related Experiment Videos
Image-type dependent eigen-regions-of-interest define conspicuity operators for predicting human scanpath fixation
Toyomi Fujita1, Claudio M Privitera, Lawrence W Stark
1Neurology and Telerobotics Units, School of Optometry, University of California Berkeley, CA 94720-2020, USA. tfujita@scan.berkeley.edu
Computers in Biology and Medicine
|August 22, 2006
Summary
This study reveals how visual perception integrates top-down information and bottom-up conspicuity. A new algorithm predicts human eye movements by analyzing visual conspicuity features from scanpath data.
Area of Science:
- Visual Perception
- Cognitive Science
- Computer Vision
Background:
- Visual perception involves both top-down (informativeness) and bottom-up (conspicuity) processing.
- An internal cognitive model guides recognition and eye movement scanpaths.
- Understanding this interplay is crucial for artificial vision systems.
Purpose of the Study:
- To define spatial visual conspicuity using eigen-features from scanpath data.
- To develop an algorithm for predicting human scanpaths based on visual conspicuity.
- To explore computer vision applications of the developed algorithm.
Main Methods:
- Utilized a self-organizing process based on principal component analysis.
- Analyzed scanpath experimental data from eight image classes.
- Defined spatial visual conspicuity from eigen-features of scanpath image loci.
Main Results:
- Demonstrated that cognitive-driven scanpath loci can be discriminated by bottom-up conspicuity features.
- Developed a conspicuity processing algorithm.
- Measured the algorithm's ability to predict human scanpaths using the positional similarity measure Sp.
Conclusions:
- Spatial visual conspicuity is a quantifiable feature derivable from scanpath data.
- The developed algorithm shows promise in predicting human visual attention.
- Findings have implications for advancing computer vision and understanding visual cognition.