Automatic Recognition of Arabic Handwritting using Hybrid Intelligent Networks

Abstract

Automatic Recognition of Arabic Handwriting is a pervasive field that has many chal- lenging complications to solve. Such complications include big databases and complex computing activities. Chapter 1 introduces our motivation and challenges, while chap- ter 2 presents our related work. Our approach is a multi-stage cascading system, and is proposed in Chapter 3. It is based on applying hybrid machine learning techniques consecutively. Hybrid cascading recognition systems aim to improve the learning ability and increase recognition rates. The approach stages start with data-mining which is essentially needed to work effectively on big databases. Agglomerative hierarchical clus- tering technique is followed to split the database into partially inter-related clusters for the data mining process. Each test image is matched to one cluster. Cluster members are then ranked in an ascending order based on our new proposed ranking algorithm. This ranking algorithm begins with computing Pyramid Histogram of oriented Gradients (PHoG), followed by measuring divergence by the Kullback-Leibler method. Finally, the classification process is applied only on the highly ranked matching classes, to assign a class membership to each test image. Adjusting the classification process to only con- sider the highly ranked database classes supports classification and enhances the overall performance.