Unit:
Κατεύθυνση / ειδίκευση Υπολογιστικά Συστήματα: Λογισμικό και Υλικό (ΣΥΣ)Πληροφορική
Supervisors info:
Σέργιος Θεοδωρίδης, Καθηγητής ΕΚΠΑ
Βασίλης Γάτος, Ερευνητής ΕΚΕΦΕ ”Δημόκριτος”
Original Title:
Exploring Character Pattern Recognition Techniques: A case study for Greek Polytonic Machine-Printed Characters
Translated title:
Exploring Character Pattern Recognition Techniques: A case study for Greek Polytonic Machine-Printed Characters
Summary:
In this thesis we explore various character pattern recognition techniques and we present a case study for Greek polytonic machine-printed characters where those techniques are applicable. We implement and describe statistical feature engineering techniques such as character zoning, adaptive character zoning, extraction of horizontal and vertical projection histograms as well as a feature extraction technique based on recursive subdivisions of the character. We also implement and discuss two classification techniques, one based on the template matching model and the other one based on artificial neural networks. Additionally, the python-based open source library that implements those functionalities is presented along with a how-to-use section. Finally, we evaluate the aforementioned techniques on two separate datasets that contain Greek polytonic characters and we present our results on the performance of our methods.
Main subject category:
Technology - Computer science
Keywords:
Optical Character Recognition, Pattern Recognition, Feature Extraction, Character Classification, Artificial Neural Networks, Greek Polytonic Characters
