A Scalable Neural Network Engine for Real-Time Optical Character Recognition

International Journal of Computer Science and Technology Vol. 2, No. 9, September 2000 | pp. 65–72

DOI: 10.46882/2000/IJCST/000021

Research Article

Title: A Scalable Neural Network Engine for Real-Time Optical Character Recognition

Names of Authors: Hans Schmidt¹, Li Wei-Dong²

Authors’ Affiliations: ¹Department of Automation Technology, University of Stuttgart, Germany; ²Institute of Microelectronics, National Taiwan University, Taipei, Taiwan.

Abstract:
Automated text recognition from high-resolution document scans demands high processing precision and real-time computational execution profiles. Traditional geometric feature extraction tools display fragile error boundaries when encountering distorted fonts, handwriting variations, or heavily pixelated source documents. This study presents a specialized, multilayer backpropagation neural network architecture optimized for accelerated character matrix classification. The model introduces a dynamic weight normalization layer combined with a custom spatial filter to isolate characters across varying resolutions. We trained our neural architecture on an industrial dataset comprising 60,000 multi-font alpha-numeric character templates. The proposed framework achieved a recognition precision rate of 96.5% across diverse alphabetic structures, representing a 7.8% accuracy advancement over conventional structural template-matching methods. The computational inference timeline averaged 12 milliseconds per document page, seamlessly matching the rapid speed boundaries demanded by high-volume commercial digital archives.

Keywords: Neural Networks, Pattern Recognition, Optical Character Recognition, Image Filtering, Document Digitization, Matrix Classification.

Manuscript Timeline: Received: June 09, 2000; Revised: July 18, 2000; Accepted: August 10, 2000; Published: September 22, 2000.

Citation: Schmidt, H., & Wei-Dong, L. (2000). A Scalable Neural Network Engine for Real-Time Optical Character Recognition. International Journal of Computer Science and Technology, 2(9), 65–72.

Subscribe to read the full article: https://www.internationalscholarsjournals.org/subscribe-to-read