Machine Reading of Arabic Manuscripts using KNN and SVM Classifiers

In this paper, feature extraction techniques like Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP) have been applied for feature extraction based on the structure of the Arabic Manuscript texts. Further, the two most popular classifiers namely, K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) have been used to classify these texts and the results of both the classifiers have been compared. For comparison, we have used 1155 images of Arabic words and 1100 Kufic words for the purpose of training and testing and then the results of both the classifiers have been compared. In this work, we have used the Arabic AHDB dataset and KUFIC dataset for experimentation and for selecting, training and testing the data; the partition approach has been used. It is found that SVM performs better than KNN and achieved maximum recognition accuracy of 97.05% (with KUFIC dataset) and 97.80% (with AHDB dataset).