CRAMM: Character recognition aided by mathematical morphology

A Character recognition system takes input in the form of scanned images of handwritten, printed or typewritten text and outputs some form of machine editable text. Intelligent Omni font CR systems have high degree of accuracy and are capable of producing formatted output that closely resembles the original input image. In this paper, a set of prominent structural features are extracted to precisely distinguish one character from the other. The classification process makes use of a decision tree classifier where at each node the decision rules are defined by some morphological operations till the final realization is done. The decision trees have been optimized for performance based on classification algorithms. The results obtained are prominent and the accuracy of our CR system is on an average 95% for handwritten text and for printed text, it achieves an accuracy of 99%. The recognition is affine transformation invariant with the assumption that individual characters are not overlapping.