A Comparative Study of Local Binary Pattern Fingerprint and Deep Learning Multi-Stage Pipeline in Document Authentication

Document forgery remains a growing global threat to institutional integrity and security. This comparative study explores two prominent approaches to document authentication: the traditional Local Binary Pattern (LBP) fingerprinting technique and a Deep Learning multi-stage pipeline. A systematic literature review methodology was adopted, analyzing key metrics such as accuracy, robustness, computational efficiency, and scalability across both models. Findings indicate that LBP methods perform efficiently under pristine conditions and are advantageous in low-resource environments due to their minimal computational demands. Conversely, deep learning pipelines demonstrated significantly higher accuracy—97.5% for closed-set and 79% for open-set recognition—along with greater resilience against noise, document damage, and variable acquisition conditions. However, these benefits come with higher resource requirements. The analysis reveals a trade-off between efficiency and accuracy, underscoring the need to align solution choice with deployment context. In high - security or large - scale applications, deep learning is preferable, while LBP remains suitable for lightweight verification tasks. Hybrid models leveraging both approaches may offer an optimized balance for real-world deployment.