A model of character recognition and legibility

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Human-Computer Interaction

Background:

• Character recognition models are crucial for understanding human perception and developing assistive technologies.
• Existing models often struggle to account for variations in legibility across different character sets and sensory modalities.
• Investigating the interplay between stimulus properties and internal representations is key to improving recognition models.

Purpose of the Study:

• To develop and validate a computational model of character recognition applicable to both visual and tactile sensory inputs.
• To explain variations in character legibility across different character sets (type, size, number).
• To account for confusions within a given character set's confusion matrix.

Main Methods:

• A two-stage processing model was developed: stimulus transformation (linear filtering, nonlinear compression) and recognition (template matching, Luce's choice model).
• The model was tested using foveal viewing of blurred/unblurred characters and tactile sensing of raised characters.
• Model performance was evaluated based on its ability to predict legibility differences and confusion matrix details.

Main Results:

• The model successfully accounts for significant differences in legibility across character sets varying in type, size, and quantity.
• It demonstrates good predictive power for overall legibility trends.
• The model shows moderate success in explaining the specific confusions within a character set's confusion matrix.

Conclusions:

• The proposed model provides a robust framework for understanding character recognition across visual and tactile modalities.
• It highlights the importance of stimulus transformation and template matching in perceptual recognition.
• Further refinement is needed to fully capture the nuances of confusion matrices in character recognition.