Related Experiment Videos
A model of attention-guided visual perception and recognition
I A Rybak1, V I Gusakova, A V Golovan
1E. I. du Pont de Nemours and Co., Central Research Department, Wilmington, DE 19880-0328, USA. rybaki@eplrx7.es.dupont.com
Vision Research
|November 3, 1998
Summary
This study presents a novel visual perception and recognition model. It effectively recognizes complex images, invariant to shifts, rotations, and scale changes, by using programmed attention and memory structures.
Area of Science:
- Computational Neuroscience
- Cognitive Science
- Computer Vision
Background:
- Understanding visual perception and recognition is crucial for AI and neuroscience.
- Existing models often struggle with invariant recognition of complex images.
Purpose of the Study:
- To introduce a new computational model for visual perception and recognition.
- To demonstrate invariant recognition capabilities for complex visual stimuli.
Main Methods:
- A two-subsystem model: low-level (fovea-like transformation, edge detection) and high-level ('what'/'where' memory structures).
- Image recognition via a 'behavioral recognition program' using programmed attention and memory-based predictions.
- Testing with complex images, including faces, under various transformations (shift, rotation, scale).
Main Results:
- The model successfully performs fovea-like transformations and edge detection.
- Separated sensory ('what') and motor ('where') memory structures are implemented.
- Invariant recognition of complex images (e.g., faces) was achieved despite transformations.
Conclusions:
- The proposed model offers a robust framework for visual perception and recognition.
- The 'behavioral recognition program' approach enables invariant image recognition.
- This model advances understanding of how visual information is processed and recognized invariantly.