Writer Identification from Offline Handwriting Images in Urdu Script with Dense-Net: A Deep Learning Approach

Handwriting serves as a valuable tool in various organizational processes, such as recruitment, interviewing, team-building, counseling, and career planning. Even identical twins with identical genes exhibit distinct handwriting styles. Leveraging this uniqueness, systems that can classify handwriting can determine the identity, age, handedness, nationality, and other details of the writer, making it an effective and reliable indicator of personality and behavior. Writer identification systems, for instance, can identify writers based on their distinct handwriting patterns. Manual extraction of handwriting features from a large dataset was once nearly impossible, but thanks to deep learning algorithms, it is now feasible to automatically extract features from documents and accurately predict the author's identity. However, there is a dearth of such systems for Urdu script, as evidenced by the literature survey. Therefore, in this paper, we propose a CNN-based model, Dense-Net, as a writer identification system trained and tested on an offline Urdu handwritten dataset contributed by 418 Urdu writers. Our model achieved an impressive overall accuracy of 99.20%, outperforming all existing methods for writer identification systems.