Optical Character Recognition of Image of Cuneiform Characters (Ugarit Alphabet)

Abstract

The research aims to provide a new contribution to the process of developing an effective algorithm for the reading Ugaritic manuscripts based on the extraction of shape and geometric features of the character. At the beginning, the text image is pre-processed to achieve the least possible noise and the highest resolution in the image to adjust the gray levels by adaptive histogram equalization. Second, the image is transformed into binary image using the dynamic Utso algorithm. The text is then segmented into its constituent lines and then the characters of each line, depending on the morphological processes, followed by the stage of extracting the geometric and morphological features of each of the segmented characters and then identifying the most important features. Before the recognition process, there is still a very important step which is the training stage forms the basis of the subsequent identification process, in which a database is produced for different characters forms in different forms and situations.

Database are stored for use the test phase.

In the last stage, the text that is stored as a test image is recognized. It is entered into the computer program. The image is processed with the same image processing steps used before. The step is the segmentation of lines and then basic characters of the image.  Features of each character are then extracted and compared with the features of all the models in the pre-stored database, Their similarities are tested based on several suggested classifiers: the Neural Network Categorizer, the SVM classifier, and the Minimal Distance Classifier (MDC) to select the character with the greatest similarity value.

After Segmentation process, we obtained 270 cuneiform characters and the correct segmentation rate was 97.4%. The system was tested on nine different scenarios and the recognition rate was 93.33% with an average time of 0.186 seconds.