Comparison of SVM and ANN performance for handwritten character classification

Kahraman F., Çapar A., Ayvaci A., Demirel H., Gokmen M.

IEEE 12th Signal Processing and Communications Applications Conference, Kusadasi, Türkiye, 28 - 30 Nisan 2004, ss.615-618 identifier identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası:
  • Doi Numarası: 10.1109/siu.2004.1338604
  • Basıldığı Şehir: Kusadasi
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.615-618
  • İstanbul Teknik Üniversitesi Adresli: Hayır

Özet

This study is about the selection of classifiers in handwritten character recogntition. The aim of the study is to determine the most appropriate classifier type for a given handwritten character feature vector. PCA based features were classified by both Multilayer Artificial Neural Networks (ANN) and Support Vector Machines (SVM), than the recognition results were compared. We select Error Backpropagation, Resilient Backpropagation and Scaled Conjugate Gradients as ANN training methods, besides selected SVM kernel types are lineer, RBF and polynomial. The experimental results shows us the SVM has beter train and test performance with respect to ANN.