Publication: Symbol Matching and character recognition
Abstract
Optik Karakter Tanima (OKT), Motif Tanima alaninda günümüzdeki baslica konulardan biridir. Bu konu, bilgisayar bilimleri arasinda imaj isleme, yapay sinir aglari, bulanik mantik, özellik çikartma ve motif tanima, ayrica matematik ve fizikten birkaç konuyu kapsayan, disiplinler arasi bir alana girer. Bugüne kadar, bu konular kullanilarak karakter tanima için pek çok teknik gelistirilmistir. OKT'lerin bazilari bu tekniklerden birini kullanirken digerleri birkaçini biraraya getirerek hibrit sistemler kurmustur. Bu proje, Birden Fazla Eksende Geçislerin Taranmasi isimli yeni bir özellik çikartma algoritmasini ve alisilagelmis puanlama tekniklerini sunar. Önerilen sistemde, yapay sinir aglarini veya bulanik mantigi kullanmak gerekli degildir. Bu özelligin kullanimi, özellik çikartma konusunda gerçeklestirilmis diger tüm özelliklerden daha iyi bir performans orani verir. Ek olarak, bir tanimlayici içinde yalniz basina kullanilabilir. Simdiye kadar gelistirilmis özelliklerden hiçbirisi, karakterleri tek basina ayirt edememektedir. Sistem, modüler bir yapiya sahiptir. Öncül ve temel adimlarin her biri, veri dosyalari üzerinden haberlesen ayri moduller halinde yapilmistir. Bu proje Borland C++ Builder 3.00 platformunda gelistirilmistir.
Optical Character Recognition (OCR) is one of the major concepts in pattern recognition today. It is an interdisciplinary study, which is based on image processing, artificial neural networks, fuzzy logic, feature extraction and pattern recognition in computer sciences also on several concepts from mathematics and physics. Many techniques have been developed for character recognition by using these concepts up to today. Several of OCRs have used one of these techniques specifically as others have used a mixture of them to form hybrid systems. This project presents an OCR, which uses a new feature extraction algorithm called Transition Scanning In Multiple Axes (TSIMA) and the traditional scoring techniques. It is not necessary to use artificial neural networks or fuzzy logic in proposed system. Use of this feature gives us a better performance ratio than all the other features implemented in the concept of feature extraction. In addition, it can be used alone in a recognizer. None of the other features developed so far, can differentiate the characters alone by itself. The system has a modular structure. Each of preprocessing and basic steps has been implemented as a single module, which communicates through data files. The project has been built in the platform of Borland C++ Builder Version 3.00. I I
Optical Character Recognition (OCR) is one of the major concepts in pattern recognition today. It is an interdisciplinary study, which is based on image processing, artificial neural networks, fuzzy logic, feature extraction and pattern recognition in computer sciences also on several concepts from mathematics and physics. Many techniques have been developed for character recognition by using these concepts up to today. Several of OCRs have used one of these techniques specifically as others have used a mixture of them to form hybrid systems. This project presents an OCR, which uses a new feature extraction algorithm called Transition Scanning In Multiple Axes (TSIMA) and the traditional scoring techniques. It is not necessary to use artificial neural networks or fuzzy logic in proposed system. Use of this feature gives us a better performance ratio than all the other features implemented in the concept of feature extraction. In addition, it can be used alone in a recognizer. None of the other features developed so far, can differentiate the characters alone by itself. The system has a modular structure. Each of preprocessing and basic steps has been implemented as a single module, which communicates through data files. The project has been built in the platform of Borland C++ Builder Version 3.00. I I