Hidden Markov models combining discrete symbols and continuous attributes in handwriting recognition

Area of Science:

• Computer Science
• Artificial Intelligence
• Pattern Recognition

Background:

• Existing handwritten word recognition models typically use either discrete or continuous features, limiting their comprehensive understanding.
• A gap exists in models that can effectively integrate both symbolic (discrete) and attribute-based (continuous) information for handwriting analysis.

Purpose of the Study:

• To develop a novel handwritten word recognition model that synergistically combines discrete symbols and continuous attributes.
• To introduce a structural feature representation for handwriting that leverages both feature types.
• To validate the proposed model's effectiveness through rigorous experimentation.

Main Methods:

• Development of structural handwriting features integrating discrete symbols and continuous attributes.
• Modeling the combined features using transition-emitting and state-emitting hidden Markov models (HMMs).
• Rigorous mathematical definition and experimental validation of the proposed HMM framework.

Main Results:

• The integrated model demonstrated superior performance compared to models relying on single feature types.
• The structural feature representation effectively captured complex handwriting characteristics.
• The hidden Markov model framework provided a robust mechanism for recognition.

Conclusions:

• Combining discrete symbols and continuous attributes in a structural model significantly enhances handwritten word recognition.
• The proposed transition-emitting and state-emitting HMMs offer a powerful approach for integrated feature modeling.
• This research advances the field of pattern recognition by providing a more comprehensive handwriting analysis method.