Invariant features for HMM based on-line handwriting recognition
In this paper we address the problem variability in handwriting due to geometric distortion of letters and words by rotation, scale and translation. In general, translation has not been a problem because it is easy to choose features that are invariant with respect to translation. It is more difficult to find features that are invariant with respect to all three types of geometric distortion. We introduce two new features for HMM based handwriting recognition that are invariant with respect to translation, rotation and scale changes. These are termed ratio of tangents and normalized curvature. Writer-independent recognition error in our system is reduced by a factor of over 50% by employing these features.
KeywordsInvariant Feature Geometric Distortion Handwriting Recognition Tangent Slope Smooth Planar Curve
- 1.Bruckstein, A. M., Holt, R. J., Netravali, A. N., Richardson, T. J.: Invariant signatures for planar shape recognition under partial occlusion. CVGIP: Image Understanding 58 (1993) 49–65Google Scholar
- 2.Hu, J., Brown, M. K., Turin, W.: Handwriting recognition with Hidden Markov Models and grammatical constraints. Fourth Int. Workshop on Frontiers in Handwriting Recognition, Taipei, Taiwan (1994) 195–205Google Scholar
- 3.Lyche, T., Schumaker, L. L., Computation of smoothing and interpolating natural splines via local bases. SIAM J. Numer. Anal. 10(6) (1973) 1027–1038Google Scholar
- 4.Boie, R., Ruedisueli, L., Wagner E.: Capacitive position sensor. U.S. patent filed (1993)Google Scholar