Handwriting Arabic Words Recognition in KHATT Dataset Based on Faster R-CNN
May Mowaffaq AL-Taee, Sonia Ben Hassen Neji, Mondher Frikha · 2023
Arabic handwriting recognition is challenging because of different writing styles and the shape of a character that changes depending on where it is in a word. Researchers always seek to propose a new state-of-the-art algorithm to address these problems and find better results for detecting and recognizing Arabic text written in handwriting. In this paper, we suggest two models that use deep learning techniques to find and classify the words in a handwritten Arabic text using a faster regional convolutional neural network (Faster-RCNN). This technique embraces three main stages. The initial stage involves preparing and annotating the dataset to identify regions of interest. The subsequent phase focuses on calculating features, while the final stage employs classification and regression layers to classify and localize words into distinct classes. In the feature extraction phase, the first model used VGG16, while the second used ResNet50. To assess the performance of the proposed method, we used the standard KHATT dataset, which has challenges in writing style, font size, and noise effects. To evaluate the accuracy and efficiency of the two models, we used different scales recall, precision, and F1_Score. Experimental evaluations achieved an accuracy rate of 99% and 98% for two networks, VGG16 and ResNet50, respectively.