M.Sc Tezi Görüntüleme

      For decades, scientists have worked on the solutions of the character recognition problem. But there are still some difficulties in this problem, so institutes and research centers continue working on this subject. Implemented solutions are based on mathematical and/or algorithmic foundations that need preprocessing of the character image such as scaling, thinning and filtering. Thus the methods which gives the results faster (i.e. real-time) that do not need preprocessing are still desired. Modeling the human beings' recognition is the most important. For this purpose, in this study an algorithmic technique that is based on the center of gravity of the image, and an heuristic technique which is based on the unification model are implemented. Because of the low fault tolerance rates of both of these methods, a third model, called unification model with fuzzy classification, is designed based on heuristic fuzziness. By defining the information base with rules and comparison criteria, the observed experimental results show that this model is closer to the real recognition model of human beings. For the characters of the same font, but with different sizes this model has a recognition rate of 98%. Also, experiments show that the correct character is one of the top two candidate characters (rates according to the defined criteria) 100% of the time.

      In this study, artificial intelligence, pattern recognition, set theory, fuzzy set theory and computer science disciplines are used.

      Keywords: Printed and Handwritten Character Recognition, Heuristic Fuzziness, Unification Model, Information Base, Feature Vector