Related Experiment Videos
A template-matching pandemonium recognizes unconstrained handwritten characters with high accuracy
1Department of Psychology, University of Copenhagen, Denmark. axel@axp.psl.ku.dk
Memory & Cognition
|March 1, 1996
Summary
This study introduces a novel visual pattern recognition model combining template matching and feature analysis. The system achieved 95.3% accuracy in recognizing handwritten digits, approaching human-level performance.
Area of Science:
- Computer Science
- Cognitive Science
- Artificial Intelligence
Background:
- Psychological research indicates internal representations like mental images aid visual pattern recognition.
- Traditional template matching methods struggle with high-accuracy recognition of complex real-life patterns, such as handwritten characters.
Purpose of the Study:
- To develop and evaluate a hybrid model for visual pattern recognition that integrates template matching with feature analysis.
- To assess the performance of this novel model in recognizing unconstrained handwritten digits.
Main Methods:
- A 'pandemonium' model was developed, utilizing multiple analyzers (demons) that compute the match between input characters and stored templates.
- Each analyzer contributes weighted evidence for character classification.
- The system was trained on unconstrained handwritten digits using an average of 37 templates per digit category.
Main Results:
- The developed system achieved a high recognition rate of 95.3% for handwritten digits.
- This performance is comparable to human recognition capabilities, falling short by only 2%-3%.
Conclusions:
- A combined template-matching and feature-analysis approach offers a robust method for visual pattern recognition.
- The 'pandemonium' model demonstrates significant potential for accurate handwritten character recognition, nearing human-level performance.