Hidden Markov models applied to on-line handwritten isolated character recognition
1Telecommun. and Traffic-Control Syst. Group, Delft Univ. of Technol.
Summary
Hidden Markov models (HMMs) effectively model handwritten character generation. This approach achieved a 6.9% average error rate for classifying lowercase English letters.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Handwritten character recognition is a challenging task in pattern recognition.
- Accurate modeling of character generation is crucial for classification.
Purpose of the Study:
- To apply Hidden Markov Models (HMMs) for modeling handwritten, isolated character generation.
- To develop a robust classification system for unknown characters using trained HMMs.
Main Methods:
- Utilized Hidden Markov Models (HMMs) to represent character generation processes.
- Employed the Baum-Welch algorithm for optimizing HMM parameters based on training data.
- Implemented maximum-likelihood classification for identifying unknown characters.
Main Results:
- Trained HMMs on examples of handwritten characters.
- Achieved an average error rate of 6.9% on the lowercase English alphabet.
- Demonstrated the effectiveness of HMMs in handwritten character classification.
Conclusions:
- Hidden Markov Models provide a viable method for modeling and classifying handwritten characters.
- The Baum-Welch optimization and maximum-likelihood classification are effective components of this system.
- The achieved error rate indicates practical applicability for character recognition tasks.
