Automated Document Labeling Using Integrated Image and Neural Processing

As part of our effort to develop an automated data entry system to identify and convert bibliographic information from paper-based documents to electronic format for inclusion in the MEDLINE database used worldwide by biomedical researchers and clinicians, we have implemented a new technique for automatically labeling zones from scanned images with meaningful labels such as title, author, affiliation, and abstract using integrated image and neural processing. Using a commercial 5-engine OCR system, scanned binary document images are first segmented into regular text zones. Each text zone is then processed to deliver an OCR output (including zone coordinates, text line information, characters and their bounding boxes, confidence levels, font sizes, and certain style attributes). From this output, features for each zone are calculated, normalized and then fed into the input layer of a back-propagation neural network for label classification. Experiments carried out on a variety of medical journals show the feasibility of using the neural network approach for label classification. Preliminary evaluation results on a sample size of several thousand images of medical journal pages show that the system is capable of labeling text zones at a classification accuracy of 98%.