Related Experiment Videos
Higher-contrast is preferred to equal-contrast in stereo-matching
1University Laboratory of Physiology, Parks Road, Oxford OX1 3PT, UK. yury.petrov@physiol.ox.ac.uk
Vision Research
|February 18, 2004
Summary
The brain prefers higher-contrast images for stereoscopic vision, challenging current stereo-matching algorithms. This study proposes a new correlation measure based on feature space scalar products, potentially explainable by primary visual cortex cells.
Area of Science:
- Neuroscience
- Computer Vision
- Computational Neuroscience
Background:
- Stereoscopic vision relies on matching images from two eyes.
- Image matching is robust to contrast differences.
Purpose of the Study:
- To investigate the preference for image contrast in global stereo matching.
- To propose a new correlation measure for stereo matching algorithms.
Main Methods:
- Presented subjects with choices between same-contrast and higher-contrast matches.
- Developed a novel correlation measure based on scalar products in multi-dimensional feature space.
Main Results:
- A higher-contrast match was consistently preferred over a same-contrast match.
- This preference contradicts traditional correlation measures that minimize image differences.
Conclusions:
- The findings challenge existing stereo-matching algorithms that minimize differences.
- A new correlation measure, maximizing scalar products in feature space, better explains human stereo vision.
- This measure is potentially computable using neural mechanisms in the primary visual cortex.