Urdu Character Recognition using Principal Component Analysis

Abstract

This paper proposes a method for Urdu language text search in image based Urdu Text. In the proposed method two databases of images have been created; first one for training purpose and another for testing purpose. Training database is named 'TrainDatabase' and testing database as 'TestDatabase'. Training database consists of all characters of Urdu language in different shapes. Eigen values and Eigen vectors of all the images to be placed in the TrainingDatabase are calculated. Only those values having highest Eigen values are kept. A feature vector for each image of the TrainDatabase is calculated by the algorithm. A threshold value is chosen such that it defines maximum allowable distance between TrainDatabase and TestDatabase images. Feature vector is also created for each image to be identified and placed in 'TestDatabase'. Comparison is done for a character to be identified with each image of 'TrainDatabase'. If the character to be recognized is matching with any character of the TrainDatabase result is shown by algorithm. MATLAB has been used as a simulation tool and the recognition rate obtained was 96.2 % for isolated characters.