WORKFLOW, BUSINESS PROCESS MANAGEMENT AND THE

Workflow, Business Process Management and the management of unstructured documents

Workflow, Business Process Management and the management of unstructured documents

Abstract

The majority of business documents (e.g., invoices, purchase orders, resumes, work orders) are arbitrarily structured and, as so, cannot be processed using conventional “ICR-friendly” document processing approaches. Alternatively, they may be consistently formatted but contain incremental variations from one document to another (e.g., medical claims, IRS forms) that diminish the effectiveness of using customary, ICR-based forms processing applications to the point where they hit unacceptable cost-justification levels. Over the past few years however, advances in automated forms processing technology have dramatically improved recognition accuracy in processing arbitrarily structured business forms. These advances include exponential increases in computing speed and memory, significant improvements in image processing technology, and innovations in neural network algorithms.

Clearly, ICR-based forms processing has come a long way since its introduction to the imaging world in the late eighties. In those early days of automated forms processing, automatically identifying a given form type and then setting it up for ICR/OCR taxed the capabilities of users, integrators, and vendors alike. Typically, the only forms eligible for image-based forms automation were those that had been created by the company responsible for processing the form. In fact, the forms in the early days were designed explicitly to be “ICR-friendly;” that is, they were specifically formatted to fit the needs of intelligent character recognition software, in order to obtain the most accurate text recognition and hand-printed recognition results possible. This meant that, preferably, the form data were printed in “drop-out” ink—carbonless ink designed to be ignored by an imaging camera—so that only the “active data” filled in by the customer was actually detected by the scanner. Furthermore, since ICR engines encounter extreme difficulty when forced to recognize connected characters, the form data fields had to be framed by graphical objects known as “combs”—strings of boxes that forced a person to separate hand-printed characters when filling out the form. The user created a software template to match each form type and define the ICR parameters of each of the fields on the form—check box, hand print, machine print, alpha, numeric, number of characters, and so forth. Due to the limitations of form identification technology, forms were processed homogeneously by the batch, carefully sorted and tightly organized by form type: one batch, one form type, so that the form template precisely fit each and every form in the batch. If the layout of any form in the batch differed incrementally from that of the form definition template, the system rejected the form and sent it to a human for manual data entry. Over the years, form identification and data location algorithms have improved considerably. Setting up a form type— or a number of form types—for automated recognition has become a relatively simple task for users to accomplish with about any given forms processing software system. An enduser can establish form ID and data field parameters in a few simple steps, ...