PLATO: Portable Language-Independent Adaptive Translation from OCR
Prem Natarajan · 2008
This is the second R&D quarterly progress report (QPR) of the BBN-led team under DARPA's MADCAT program. The goal for the pre-processing and image enhancement task is to eliminate noise artifacts from documents. In this reporting period, we performed preliminary experiments to assess the usefulness of shape-DNA enhancement on machine-print and handwritten images. The shape-DNA approach uses a database of low- and high-resolution shapes and a probabilistic shape-mapping model. The database and mapping are both automatically learned from training data to estimate high-resolution details from low-resolution shapes.