Abstract
Mobile devices with high-resolution cameras and computational capabilities have become ubiquitous tools for document digitization. However, existing on-device OCR solutions primarily focus on text recognition, lacking the ability to capture document structure or extract semantic context. This paper introduces a hybrid mobile-cloud architecture for Intelligent Document Processing (IDP) that combines real-time on-device document capture with cloud-based OCR, document classification, and semantic information extraction. Smartphones function as intelligent sensing nodes, performing image enhancement, perspective correction, keyframe selection, and secure transmission to a cloud backend. The backend employs Transformer-based classifiers and multi-modal language models to extract structured data, including dates, financial amounts, and reference numbers, supporting multiple languages and document types. Experiments on real-world invoices demonstrate that the hybrid system reduces end-to-end processing time compared to cloud-only approaches while maintaining high extraction accuracy, even under variable network conditions. Video capture further enhances reliability by selecting optimal frames. The architecture also addresses practical enterprise needs, including data security, compliance, and integration with ERP, CRM, and document management systems. Future work will explore lightweight on-device semantic pre-processing, expanded document type support, and adaptive transmission strategies to improve efficiency and resilience in large-scale enterprise applications.