Combining different off-line handwritten character recognizers

Abstract

This present work presents a recognizer based on the combination of three Support Vector Machine (SVM) classifiers whose inputs have different parameters from characters. The three approaches of feature extraction for handwritten off-line digits, capital letters and lower case letters, have been chosen for improving the combination using database NIST-SD19. We have applied pre-processing in order to achieve greater inter-class discrimination and similarity. These three feature extractions are based on Kirsch masks with and without slant correction and Fourier elliptic descriptors.