Digit Recognition
Manolis Kamvysselis

I. Introduction
  • Abstract
  • Algorithm Overview
  • Organization of the paper
    II. Working with online data
  • Relation with Offline Data
  • Lower-dimensional online data
    III. Character transformation to Angle Derivative Space
  • Angle derivative space
  • Motivation for representation
  • Resolving angle ambiguities
  • The loop ambiguity
  • Working with Periodic Costs
  • Enriching the representation
    IV. Noise Reduction
  • Fourier methods for denoising
  • Brute force compression
  • Wavelets bring out curve features
    V. Feature Selection
    VI. Supervised Learning
  • Training gaussian mixtures
  • Evaluating new data
    VII. Results
  • Good performance for long sequences
  • Individual character results
    VIII. Applications / Future work
  • New characters from a distribution
  • Handwriting style recognition
  • Multi-dimensional data
  • Wavelets tailored to characters
    IX. Unsupervised Learning
  • Extension to offline data
    X. Conclusion
    About this document ...
    Manolis Kamvysselis
    2000-06-11