Using Faster R-CNN to Detect and Recognize Arabic Handwritten Words

Abstract

Arabic handwriting recognition is a crucial area of computer vision research. Still, its complexity, diverse writing styles, and overlapping words have led to a lack of published research in this field. This paper suggests two new models to recognize handwritten Arabic words, depending on the Faster Region-Convolution Neural Network (Faster R-CNN). These models used two pre-trained networks during the feature extraction phase: The Visual Geometry Group-16 (VGG-16) network and the Residual Network (ResNet50) network. Models are independently trained and tested on two datasets: The Institut Für Nachrichtentechnik/Ecole Nationale d'Ingénieurs de Tunis (IFN/ENIT) dataset and the KFUPM Handwritten Arabic Text (KHATT) dataset. Test results showed that the proposed models give excellent results compared to others. The results of VGG16 and ResNet50 with the IFN/ENIT dataset reached accuracy rates of 92% and 100%, respectively. Meanwhile, the accuracy of the KHATT dataset reached 99.4% and 98% with VGG16 and ResNet50, respectively.