In this thesis, a classifier based on local sub-unit level and global character level representations of a character, using stroke direction and order variations independent features, is developed for recognition of ...