Linear dimension reduction methods in character recognition systems

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

IEEE 12th Signal Processing and Communications Applications Conference, Kusadasi, Turkey, 28 - 30 April 2004, pp.611-614 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu.2004.1338603
  • City: Kusadasi
  • Country: Turkey
  • Page Numbers: pp.611-614
  • Istanbul Technical University Affiliated: No


Handwritten Character Recognition Systems can be divided into three steps: pre-processing, feature extraction and classification. In the feature extraction process, representation power of features should be increased yet keeping the number of features as small as possible. In this study, raw character image vectors are projected to lower dimension spaces by different linear transformations and their representation and discrimination power are compared. In dimension reduction, Principal Component Analysis (PCA), Multiple Discriminant Analysis (MDA) and Independent Component Analysis (ICA) are compared and best classification performance is obtained by using ICA. A multi layer perceptron, which is trained by conjugate gradient algorithm, is used for classification. The handwritten character database, studied on, consists of 5000 training patterns and 2500 test patterns.