Processing and Recognition of Handwritten Documents

Nowadays, the accurate recognition of machine printed characters is considered largely a solved problem. A lot of commercial products are focused towards that direction, achieving high recognition rates. However, handwritten character recognition is comparatively difficult. So, the recognition of handwritten documents is still a subject of active research. In this thesis we studied the processing and focused on the recognition stages for handwritten optical character recognition. At the recognition stage a feature vector is extracted for all extracted characters in order to classify them to predefined classes using machine learning techniques. We studied several feature extraction techniques and developed methodologies that efficiently combine different types of features. Furthermore, a novel methodology that extracts features and classifies characters using a hierarchical scheme is proposed. This methodology, after being tested on wellknown character databases, as well as on databases consisting of characters from historical documents and a database consisting of Greek contemporary handwritten characters, that were particularly created in this thesis, achieved recognition rates that are among the best one can find in the literature. This methodology was also applied to cursive handwritten words. The recognition rates in these experiments were also very high. Finally, an algorithm that automatically estimates the free parameters involved in character segmentation is also suggested. Character segmentation is very important because its result affects directly the recognition rates. Thus, the optimal segmentation is essential for a successful recognition.