Optical Character Recognition for Urdu Text: A Review of Techniques, Challenges, and Future Directions

In recent years, Urdu Optical Character Recognition (OCR) systems have received considerable attention due to the increasing demand for automated recognition and analysis of text, which requires the development of robust OCR systems to convert printed and handwritten Urdu text into digital format. This article aims to provide a comprehensive overview of the most recent advances in OCR systems for Urdu. Analysing and synthesising 30 relevant research articles published between 2018 and 2023 to determine their contributions and future directions. Numerous studies have explored the use of deep neural networks, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory networks (LSTMs), to achieve high accuracy in character recognition by incorporating evolutionary algorithms, transfer learning, augmentation, and holistic unconstrained recognition systems. These models have demonstrated promising offline and online Urdu handwriting recognition results. The reviewed articles also shed light on the difficulties and limitations of existing Urdu OCR systems, such as the lack of standardised datasets, the dearth of annotated training data, and the need for robustness against noise and variations in handwriting styles. The literature discusses the development of more efficient algorithms, the investigation of multi-domain approaches, and the integration of deep learning techniques with other image processing techniques as future directions. The findings of this review will help researchers and practitioners in understanding the most recent techniques and challenges in Urdu OCR systems.