Artificial Immune Algorithm for Handwritten Arabic Word Recognition

In this work, a system for solving handwritten Arabic word recognition is proposed. The aim is focused on holistic word recognition, which is devoted to recognize averaged size lexicons by using a single classifier. Presently, we investigate the applicability of the Artificial Immune Recognition System (AIRS) to achieve the recognition task. For the feature generation step, ridgelet transform and pixel density features are combined to highlight both linear singularities and topological traits of Arabic words. Experiments are conducted on a vocabulary of twenty-four words extracted from the IFN/ENIT dataset. The results show that feature combination improves the recognition accuracy with more than 1%. The comparison with Support Vector Machine (SVM) classifier highlights the effectiveness of AIRS. This latter achieves comparable and sometimes better performance than SVM and can be extended to recognize any number of classes.