Autonomous Intelligent Document Analysis system (AIDA)

Optical Character Recognition (OCR) is critical to document management workflow in both Army and civilian scenarios. It is also one of its weakest links, having no turnkey or standardized solution. Low quality documents imaged under adverse conditionsexhibit severe image artifacts, such as uneven illumination, low contrast, speckle noise, degraded character glyphs (broken, joined, aliased), rotational skew, and page warp (perspective skew and curved baselines). These artifacts catastrophically degradeOCR performance. Processing large volumes of such documents using manual clean-up is prohibitively expensive.We propose to enhance and fully implement our Autonomous Intelligent Document Analysis system (AIDA), which pre-processes document images (bitmaps) to boost OCR performance. AIDA consists of a set of document image quality metrics, each detecting aspecific artifact type, and a suite of corresponding image enhancement algorithms. AIDA's greatest novelty is a knowledge-based approach to automatically determine the optimal enhancement operator sequence and parameter values, based solely on computedquality metrics. We demonstrated feasibility in Phase I by developing detection and correction of three artifact types in Arabic and English documents and by prototyping the knowledge-base framework. In Phase II, we will handle character glyph restoration,tackle additional artifacts that severely degrade performance, and explore multi-OCR optimization.AIDA can have an immediate role in increasing OCR accuracy for anti-terrorist intelligence (for example in the CHIMS Army project), where documents are frequently imaged under adverse conditions. AIDA also has direct applicability to financial services,medical records, litigation support, and record archiving for corporations, universities, and government. Envisioned product packaging includes stand-alone software and software plug-ins for existing OCR systems.