A septon feature scheme in handwritten digit recognition

Author

Agba, Lawrence C. ; Shankar, Ravi

Author_Institution

Div. of Sci. & Math., Bethune-Cookman Coll., Daytona Beach, FL, USA

Volume

3

fYear

1992

fDate

7-11 Jun 1992

Firstpage

608

Abstract

A set of comprehensive features suitable for character recognition is proposed. The features are extracted from thinned binary images and classified using an artificial neural network. A special application of this septon feature extraction scheme in handwritten digit recognition is described. The initial experimental results on a sample of handwritten digits from the United States Postal Service database and the National Institute of Standards and Technology database are very encouraging. Of the training data set of 3811 handwritten digits, 99.95% were recognized. The system was able to recognize 91.15% of the remaining 1909-digit never-before-seen test set. Upon training with all the 5720 digits, 99.34% were recognized

Keywords

artificial intelligence; character recognition; feature extraction; neural nets; National Institute of Standards; United States Postal Service database; artificial neural network; character recognition; comprehensive features; handwritten digit recognition; septon feature scheme; thinned binary images; Artificial neural networks; Character recognition; Feature extraction; Handwriting recognition; Image databases; NIST; Postal services; Spatial databases; System testing; Training data;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks, 1992. IJCNN., International Joint Conference on

Conference_Location

Baltimore, MD

Print_ISBN

0-7803-0559-0

Type

conf

DOI

10.1109/IJCNN.1992.227107

Filename

227107