ScanSoft OmniPage Pro 12 Office

Despite all of ScanSoft's recent acquisitions, OmniPage Pro remains its flagship OCR package, with TextBridge (see opposite) as the low-cost offering. Recent new releases of OmniPage haven't been terribly exciting, but OmniPage Pro 12 is a different story altogether.

It's undoubtedly a major update and has earned itself the suffix of Office. Notable changes include improved PDF conversion - one of the few notable new features offered by OmniPage Pro 11 - enhanced accuracy and improved support for network use. There are also some better enterprise features, as well as refined proofing and added automation for unattended document processing.

ScanSoft says it has worked hard to deliver the enhancements because of the increasing use of PDF documents and multifunction devices. OmniPage Pro 12 can also leverage the Internet with its new automatic processing capabilities.

Don't expect greatly enhanced all-round accuracy, though. As OCR technology has matured, we've come to expect less dramatic improvements with new versions. A claimed 50 per cent improvement in accuracy might sound a lot, but if accuracy is already at 99 per cent - given a particular level of source document image quality - this amounts to less than 1 percentage point improvement in total.

The target for OCR development has been mainly focused on translating poor-quality originals. OmniPage Pro 12 boasts three OCR engine technologies, originating from ScanSoft acquisitions Caere, Calera and Xerox TextBridge, from which ScanSoft itself originates. A unique system of engine 'voting' optimises the OCR process, enabling the most effective engine to be used on parts of a given document.

There are also signs that ScanSoft's other recent acquisition - Lernout & Hauspie's speech-recognition technology - is influencing OmniPage's development. Not only does Pro 12 use dictionaries, including sector-specific ones that address legal and other vocabularies, but it also employs a degree of context checking. On top of that, the remarkably human-sounding, Lernout & Hauspie-developed RealSpeak Text-to-Speech voice synthesis package is bundled with OmniPage Pro 12.

Getting the words right is one thing, but ScanSoft has made some important steps to make better use of them when laying out the recognised page. Our complex layout test page showed a marked improvement in text accuracy and formatting compared to OmniPage Pro 11. Far less text was incorrectly identified as part of a graphic, and multiple columns are now linked. So if you decide to edit the document afterwards, the text flows naturally from one column to another.

ScanSoft says, with some conviction, that this is the difference between looking right and working right. OmniPage Pro 12 is also good at retaining the font size and style, as well as both text and graphic object positioning. The only thing it wouldn't attempt to do on our test page was reproduce the vertical and horizontal lines that separate subheadings, columns and boxed sections.

Incidentally, there's also a new book-processing feature that enables opposing pages to be scanned and processed at the same time.

Until we'd seen OmniPage Pro 12, ABBYY FineReader 6 Professional (see Reviews, issue 96, p138) was the clear leader in overall OCR performance. However, OmniPage Pro 12 produces a more coherent result from difficult originals, and FineReader 6 also lacks the convenience of providing text flowing from column to column.

It's interesting to see that the areas FineReader 6 had difficulty in reading correctly were often the areas in which OmniPage Pro 12 excelled, and vice versa. But on balance, OmniPage Pro 12 is the better performer in terms of accuracy and layout retention. It also implements multithreading to enable the simultaneous scanning of new pages and the recognition of previous ones. This is handy for large-volume batches, as it isn't particularly quick at processing pages during the recognition stage.