A BACKPROPAGATION NEURAL NETWORK APPROACH FOR OTTOMAN CHARACTER RECOGNITION PGNKBU

prudencewooshAI and Robotics

Oct 19, 2013 (4 years and 19 days ago)

67 views



Intelligent Automation and Soft Computing, Vol. 15, No. 3 pp. 451-462, 2009
Copyright © 2009, TSI
®
Press
Printed in the USA. All rights reserved

451

A BACKPROPAGATION NEURAL NETWORK APPROACH FOR
OTTOMAN CHARACTER RECOGNITION

P
ELIN
G
ORGEL
1

N
IYAZI
K
ILIC
2

B
IRSEN
U
CAN
3
A
HMET
K
ALA
3

O
SMAN
N.

U
CAN
2

1
Istanbul University
Engineering Faculty
Computer Science Dept. 34320 Avcilar
Istanbul, Turkey

2
Istanbul University
Engineering Faculty
Electrical & Electronics Dept. 34320 Avcilar
Istanbul, Turkey

3
Istanbul University
Economics Faculty Avcilar
Istanbul, Turkey


ABSTRACT—The Ottoman Empire established in 1299 and continued 6 centuries
covering an area of about 5.6 million squared km. The Empire left a large collection of
valuable archives interesting to historians from all over the world. Investigation and
understanding these documents will shed light on the history of the world. In order to
achieve access of the considered information by worldwide scientists, it is essential to
translate Ottoman characters into Latin alphabet. Thus, we aimed to recognize the
Ottoman characters using Artificial Neural Network (ANN) and compared it with Support
Vector Machine (SVM) approaches. We used printed type of Ottoman scripts in image
acquisition. Pre-processing such as normalization and edge detection were implemented.
Multilayer perceptions of ANN were trained using the backpropagation learning
algorithm. As a result of our research, we are able to classify the Ottoman characters with
85.5% classification accuracy using the proposed recognition system.

Key Words: Artificial neural network, Backpropagation, Ottoman script, Cellular neural
network edge detector, Character recognition