Arabic Handwriting Recognition Using Deep Learning and Simulated Annealing

Arabic handwritten characters are important in many applications, making reliable recognition systems essential. In this study, we present a new method for Arabic Handwritten Recognition (AHR) that doesn't rely on segmentation. Using Simulated Annealing, we optimized the architectures of RegNetY-32GF and NfNet-F5 deep convolutional neural networks (CNNs) for feature extraction and classification. The HACDB database was used to demonstrate the effectiveness of our approach. When compared to other methods, our approach proved to be highly effective and robust. The optimized RegNetY-32GF model achieved an impressive success rate of 98.27%, highlighting the strength of our method.