Systematic Literature Review on the Integration of Wasserstein GANs and Convolutional Neural Networks for Urdu Handwritten Character Recognition

This systematic literature review examines the integration of Wasserstein Generative Adversarial Networks (WGANs) and Convolutional Neural Networks (CNNs) for Urdu Handwritten Character Recognition (UHCR), a domain underrepresented in mainstream OCR research. Using a PRISMA-guided methodology, 25 peer-reviewed studies published between 2020 and 2024 were synthesized from seven major academic databases. The review evaluates how WGANs enhance data augmentation and reduce annotation effort, while CNNs contribute to robust classification in low-resource linguistic contexts. Six research questions guided the analysis, focusing on model effectiveness, dataset limitations, transfer learning, and evaluation metrics. Findings indicate that hybrid WGAN-CNN architectures significantly improve recognition accuracy, particularly for cursive ligatures and underrepresented Urdu characters, while mitigating challenges of manual annotation and dataset imbalance. Transfer learning from Arabic and Persian datasets further strengthens model robustness. Keyword co-occurrence analysis using VOS viewer revealed two dominant thematic clusters: one centered on biometric technologies and feature extraction, and another on machine learning and character recognition. These insights position UHCR within a broader computational and biometric research ecosystem. The review concludes by outlining future directions, including cross-lingual OCR systems, real-time handwriting recognition applications, and synthetic dataset generators for low-resource scripts.