Related Experiment Videos
A neuro-cognitive visual system for object recognition based on testing of interactive attentional top-down
1Siemens AG, Corporate Technology, ZT IK 4, Otto-Hahn-Ring 6, 81739 Munich, Germany. Gustavo.Deco@mchp.siemens.de
Perception
|February 28, 2001
Summary
This study enhances object recognition models by decoupling visual processing from programmed eye movements. The new system allows for flexible fixation sequences, improving recognition adaptability.
Area of Science:
- Neuroscience
- Computer Vision
- Cognitive Science
Background:
- Existing object recognition models often rely on programmed scanpaths, limiting flexibility.
- Primate visual system organization, with distinct 'what' and 'where' pathways, provides a foundation for biological plausibility.
Purpose of the Study:
- To extend a systemic model for object recognition based on primate visual system organization.
- To develop a more flexible learning and recognition mechanism by removing the assumption of a programmed behavioral scanpath.
Main Methods:
- The proposed model separates visual processing into 'what' (ventral) and 'where' (dorsal) subsystems.
- The 'where' system analyzes spatial relationships between successive fixations and object features, rather than programming a scanpath.
- Object recognition is achieved by identifying parts and their spatial configurations.
Main Results:
- The extended model demonstrates increased flexibility in object recognition by allowing varied fixation sequences.
- Decoupling recognition from a fixed scanpath enhances the system's adaptability to different object viewing strategies.
Conclusions:
- The proposed model offers a more flexible and potentially more biologically realistic approach to object recognition.
- This framework advances computational models of vision by integrating spatial relationships more dynamically into the recognition process.