Optical Character Recognition

1
Optical Character Recognition

• - Proposer -
• Prof. S. Rama Mohan
• Department of Applied Mathematics
• Faculty of Technology and Engineering
• The M. S. University of Baroda
• Vadodara 390002. Gujarat

Language Gujarati

• Components that will be implemented
• Page Layout Analysis Engine
• Visual Component Extraction Engine
• Visual Component Recognizer Engine
• Post Processor Engine
• Training and Testing Data

2
Page Layout Analysis Engine

• List of the technique (s) that will be used
• Smearing and Histogram Analysis (For locating
  heading (s), paragraphs, )
• Connected components Analysis. (Skew estimation,
  Text-Image Separation)
• Performance for these techniques in other
  Languages
• The claims in other languages are to be evaluated
• Estimate of the expected performance (Quarterly
  PERT Chart)
• To be worked out
• Name of the domain for which the performance will
  be optimized
• Magazine articles with regular shaped images
• Other evaluation metrics in addition to the
  domain
• Preformatted forms

3
Visual Component Extraction Engine

• List of the technique (s) that will be used
• Connected component Analysis
• Artificial Neural Networks (Unsupervised)
• Performance for these techniques in other
  Languages
• Recorded quality performance for European scripts
• Experimental approaches for Indian scripts
• Estimate of the expected performance (Quarterly
  PERT Chart)
• To be worked out
• Name of the domain for which the performance will
  be optimized
• Printed documents of computerized type set
• Other evaluation metrics in addition to the
  domain
• Printed books (robustness check against banding
  noise etc.)

4
Visual Component Recognizer Engine

• List of the technique (s) that will be used
• Performance for these techniques in other
  Languages
• Established performance in European scripts
• Template matching with Fringe Maps gives more
  than 90 character level recognition accuracy for
  Telugu script.
• Estimate of the expected performance (Quarterly
  PERT Chart)
• To be worked out
• Name of the domain for which the performance will
  be optimized
• Printed documents of computerized type set
• Other evaluation metrics in addition to the
  domain
• Printed books

5
Post Processor Engine

• List of the technique (s) that will be used
• N-gram analysis
• OCRized Dictionary and spell checker
• Confusion matrix
• Performance for these techniques in other
  Languages
• Extensively used for the European languages
• Estimate of the expected performance (Quarterly
  PERT Chart)
• To be worked out
• Name of the domain for which the performance will
  be optimized
• Magazine articles on sports.
• Other evaluation metrics in addition to the
  domain
• Government circulars

6
Training and Testing Data

• List of the technique (s) that will be used
• Ground truth data generation with information for
  every identified components
• Performance for these techniques in other
  Languages
• Encouraging performance in Telugu, Bangla and
  Devnagari
• Estimate of the expected performance (Quarterly
  PERT Chart)
• To be worked out
• Name of the domain for which the performance will
  be optimized
• Magazine articles on sports
• Other evaluation metrics in addition to the
  domain
• Government circulars