Ottoman Alphabet Character Recognition by LDA

Abstract

In our study, an Ottoman Turkish language recognition system by LDA (linear discriminant analysis) approach is improved. LDA reduces dimensionality of original data while expressing most important characteristic features. It aims maximizing "between classes differences" and minimizing "within class differences". In our system, the train set consisted of 33 classes that each class expresses each character of Ottoman Turkish language alphabet. The training set images were normalized by the following preprocessing steps; the contrast of the images was increased and images were aligned and resized. The characteristic features that used in recognition were calculated by LDA method using dimensional reduction applied images. The described processes were applied to the test set images and then the distance of test images to the classes were calculated by Euclidian distance method.