Related Experiment Video
Updated: Jul 7, 2026
06:37
Artificial Intelligence-Based System for Detecting Attention Levels in Students
Published on: December 15, 2023
Neural network-based systems for handprint OCR applications
M D Ganis1, C L Wilson, J L Blue
1Nat. Inst. of Stand. and Technol., Gaithersburg, MD 20899, USA. mgarris@nist.gov
Summary
Neural networks, specifically enhanced multilayer perceptrons, offer improved optical character recognition (OCR) for handprint data. NIST developed a system using novel neuron activation functions and regularization techniques for better performance.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Neural network (NN) approaches have significantly advanced optical character recognition (OCR) performance.
- Handprint recognition remains a challenging area within OCR applications.
Purpose of the Study:
- To present an enhanced neural network classification scheme for form-based handprint OCR.
- To describe an end-to-end OCR system developed by NIST.
Main Methods:
- Utilized an enhanced multilayer perceptron (MLP) with novel neuron activation functions.
- Implemented successive regularization and Boltzmann pruning to optimize the MLP's weight space.
- Developed an end-to-end system for form-based handprint recognition.
Main Results:
- The enhanced MLP design addresses singular Jacobians, constraining weight space volume and dimension.
- Performance characterization studies were conducted on NN systems, including the NIST system.
- Evaluations were performed at the first OCR systems conference.
Conclusions:
- The presented NN classification scheme and NIST system offer advancements in handprint OCR.
- The enhancements to the MLP contribute to improved accuracy and robustness in character recognition.
- Further performance characterization validates the effectiveness of these neural network approaches.
