Bilgisayar Mühendisliği Bölümü Tezleri

Browse

Now showing 1 - 1 of 1

Citation Count: SHERTIL, M.S. (2005). Appliying neural networks to character recognition. Yayımlanmamış yüksek lisans tezi. Ankara: Çankaya Üniversitesi Fen Bilimleri Enstitüsü.
Appliying neural networks to character recognition
(2005-06) Shertil, Mahmud Saad; Çankaya Üniversitesi, Fen Bilimleri Enstitüsü, Bilgisayar Mühendisliği Bölümü
Optical Character Recognition (OCR) is one of the most widely used applications of automatic pattern recognition and it is a very active research field since the 50's. Today there are numerous algorithms that perform this task, each with its own strengths and weaknesses. In this thesis we explore neural networks to find the best topology regarding the number of layers and the number of neurons at each layer by classifying two different datasets using backpropagation networks with two different approaches. We then compare the results between them and show that the obtained optimum set of parameters is valid for both printed letter and the handwritten digits. Moreover the letter dataset consists of sixteen features with 20000 samples and twenty-six classes. Here we see that the number of classes is more than the number of features which affects our results. The second dataset which is handwritten digits consists of sixty-four features with 5620 samples and ten classes. Here we see that the results are better than the first dataset