PDF-XChange Editor OCR: What Each Engine Produces

PDF-XChange Editor runs OCR through the OCR Pages command on its Convert tab, and it has two engines. The vendor's manual says the default engine adds an invisible text layer over the scanned page. The Enhanced OCR engine, licensed with PDF-XChange Editor Plus, adds a choice of output: Searchable Image, Editable Text and Images, or Fine Page Content.

For a legal file, that choice of output decides whether the page you file is still the page you scanned. This post follows the vendor's manual for versions 10 and 11 of the Editor and its product pages, and quotes them where the wording matters.

Does PDF-XChange Editor include OCR?

Yes. The PDF-XChange Editor manual documents a command called OCR Pages on the Convert tab, and says two optical character recognition engines are available: a default engine and an Enhanced engine. The version 11 manual describes the default engine as "available to all users" and the Enhanced engine as available only with a PDF-XChange Editor Plus licence.

The two manual pages this post relies on are the version 11 page on performing OCR on PDFs in PDF-XChange Editor and the version 10 page on the OCR Pages command. They describe the same dialog with small differences in labels, noted where they matter.

Three ways the manual says you can run it

• OCR Pages, on the Convert tab, for a whole document or a range of pages.
• OCR Selected Region. The version 11 manual says this is possible after either the Snapshot Tool or the Crop Page Tool has been used to define a page area, and that the option appears in the right-click shortcut menu.
• Choosing the engine. The version 11 manual's page on OCR preferences says you "Use the dropdown menu to select either the Enhanced or Default OCR engine." The version 10 page places that setting under Preferences on the File tab.

Things to confirm on your own installation

The vendor's product page for PDF-XChange Editor lists portable downloads labelled "No OCR" beside the ordinary ones. If the OCR Pages button is missing or does nothing, find out which build your IT department installed.

The same product page says of the free version that "70% of features are unrestricted", and it marks some features as licensed. Its tooltip for a licensed feature reads: "It is available in the free version but will add watermarks to documents when it is used." Before you run any OCR command on a client document with an unlicensed copy, run it on a blank test page and look at the output for a watermark.

What the default PDF-XChange OCR engine adds to a scanned page

The default PDF-XChange OCR engine adds an invisible layer of text over the words in the picture. The version 10 manual says the default engine "analyzes image-based documents, recognizes text and then places a duplicate, invisible text layer on top of it, which makes the source text selectable and searchable in the same manner as ordinary text."

The version 11 manual says the same thing in slightly different words: the default engine "recognizes text in image-based documents and then inserts an invisible layer of duplicate text over it, which makes the image-based text searchable and selectable."

That is the build of a searchable PDF. The page stays a picture, and the text sits where you cannot see it. The guide to how to make a scanned PDF searchable explains how that layer works in any tool.

Running it

1. Save a copy of the scan and open the copy.
2. Click the Convert tab, then click OCR Pages. The OCR Pages dialog box opens.
3. Set the page range, the recognition options and the output options.
4. Click OK.
5. Search for a word on a middle page, then select a line and copy it into a blank document.

One option that decides which file is changed

The Output Options of the default dialog contain a box that deserves attention before any other. The manual says: "Select the Create a New Document box to create a new document for the output of the optical character recognition. If this box is not selected, then the original document will be updated with the output instead."

So with the box clear, the file you have open is the one that changes. For a document that came from a client or the other side, tick the box or work on a copy. Doing both costs nothing.

What the default engine's page of the manual does not promise

The manual describes what the default engine does to the file. It gives no accuracy figure and does not say that doubtful words are marked. The manual does compare the two engines on speed and accuracy, in the Enhanced engine's favor. That is the vendor's comparison of its own products, and this post has not tested it.

What does the Enhanced OCR engine in PDF-XChange Editor Plus add?

The Enhanced OCR engine adds a choice of three output types, where the default engine has one. The manual lists Searchable Image, Editable Text and Images, and Fine Page Content. It says the Enhanced engine is available when PDF-XChange Editor Plus is purchased, as a stand-alone product or as part of the PDF-XChange PRO bundle.

The three outputs, in the manual's words

The version 10 page lists a fourth control in the Enhanced dialog, a box called Draw Lines for tables, which it says is "to replace recognized column/row lines in tables with editable vector lines in output documents." The version 11 page opened for this post does not mention that box. Look at your own dialog to see whether it is there.

The manual adds that the other options in the Enhanced dialog are the same as in the default one.

Which licence has it

The vendor's pages are consistent on this:

• The version 11 manual: the Enhanced engine "is available only with a PDF-XChange Editor Plus license."
• The product page: "PDF-XChange Editor Plus comes with an Enhanced OCR engine", and the features of Editor Plus "are not available in an unrestricted capacity with a regular PDF-XChange Editor license."
• The vendor's page for the Enhanced OCR plugin marks each of its features as licensed, and says of each: "It is available in the free version but will add watermarks to documents when it is used."

That last page lists the plugin's features by name: improved character recognition, text style recognition, table recognition, multiple output options, multi-threaded processing, automatic detection and correction of image skew and incorrect page rotation, and installing or updating OCR languages without restarting the application. Those are the vendor's labels. For current prices, go to the vendor's site.

Does a plain Editor licence need the Plus upgrade?

Not for a searchable PDF. Per the manual, the default engine's single output is the invisible text layer, and Searchable Image in the Enhanced dialog is described in nearly the same words. The Plus licence is what you need if you want the page turned into editable text inside the Editor.

Page range, accuracy and skew options in the OCR Pages dialog

The dialog has three groups of settings: Page Range, Recognition Options and Output Options. Page Range picks the pages. Recognition Options set the languages, the accuracy and what to detect or ignore. Output Options decide whether skew and rotation are fixed and whether a new document is created.

Page range

In version 10 the choices are All, Current and Custom, with a Subset of All, Odd or Even. In version 11 they are labelled All, Current Page and Pages, with a Subset dropdown of All Pages, Odd Pages Only or Even Pages Only.

The manual's page on page range settings gives the syntax for a custom range:

The caret counts from the end of the document, which is handy for a signature page whose number you do not know.

Accuracy

The manual is candid about the accuracy setting. It says "increasing the accuracy also increases the time that the process takes and vice versa", and then: "setting the accuracy to high may result in unusual output if the document contains imperfections. This is because the software will search to a greater depth and may attempt to recognize imperfections as text."

In other words, on a speckled fax or a photocopy with a dark edge, the highest setting can turn dirt into characters. If a first run leaves stray symbols in the margins when you paste the text, try a lower accuracy setting on that file and compare.

Detect and ignore

The Recognition Options include six boxes:

• Detect skew of page content, for pages the manual describes as "scanned crookedly"
• Detect incorrect page rotation
• Ignore company logos, with an ellipsis button to add, view and manage logos
• Ignore existing text on page
• Ignore comments on page
• Ignore form fields on page

The three Ignore boxes for text, comments and form fields matter for a mixed file. A PDF that is part typed pages and part scans already has real text on some pages, and the manual describes these boxes as omitting that content from recognition.

Fixing skew and rotation

Detecting is one setting and correcting is another. Under Output Options, the manual says: "Select the Fix content skew and incorrect page rotation box to deskew pages that are scanned crookedly and auto-correct page rotation issues."

Ticking that box means the output page is straightened or turned. That is a change to how the page looks, and a welcome one for a working copy. For a record copy, note that you applied it, and keep the scan as it arrived. The post on whether OCR changes the original PDF covers how to show that the two copies depict the same pages.

Adding and updating PDF-XChange OCR language packs

Languages are added from inside the program. The manual says: "Click Add/Update Languages to add/update the language packs used for OCR." The button is in the Recognition Options of the OCR Pages dialog, and the version 11 manual lists an Add/Update languages control in the OCR preferences as well.

What the vendor says about language files

The vendor's web page of OCR language files for the Editor, Tools and Viewer offers downloadable language packs grouped by region, and carries a notice that those files "are intended for the older versions" of the product and will not work with version 8 or newer. For newer versions, the page says to use the "Add/update Languages" button "presented within the OCR function of the installed Application."

So the route depends on the version:

More than one language

The version 11 manual states: "Note that multiple languages can be used simultaneously for OCR." That is the setting for a bilingual contract or an English file with exhibits in another language. Select only the languages that are on the pages. How the language choice affects a result is explained in the post on conversion options for a scanned PDF.

On a managed computer

The Add/Update Languages button downloads files. On an office machine where downloads from inside a program are blocked, the button may fail. Ask IT to add the languages you need before a deadline depends on them. The plugin page lists installing or updating OCR languages without an application restart as a feature of the Enhanced plugin, and the manual does not say whether the default engine needs one, so allow for a restart after adding a language.

When does editable-text output from PDF-XChange change what the page shows?

Editable-text output from PDF-XChange changes what the page shows whenever you choose Editable Text and Images or Fine Page Content. Per the manual, the first replaces image-based text with recognized text, and the second replaces the page's content with new content built from what was recognized. Only Searchable Image, or the default engine, keeps the picture and adds hidden text.

Why that matters for an exhibit

With a searchable image, a misread word is wrong only in the hidden layer. A search may miss it, but the page still shows what the paper showed.

With either of the other two outputs, the recognized text is what appears on the page. A "3" read as an "8" is now printed on the page as an "8". The two differ in what they keep. The version 10 manual says Editable Text and Images will "retain existing content such as text and images" while it converts image-based text into editable text. It says of Fine Page Content that the new content "contains only the text and images recognized", which tells you that anything not recognized is not carried over. Neither description says what happens to a signature, a stamp or a handwritten initial, so compare those marks with the scan after either output.

The manual also notes that where a page is one large graphic with some text zones over it, the visual output of Editable Text and Images and of Fine Page Content "will be very similar". The vendor does not say they will match the scan.

A rule for the file

Take a fictional case. Nadia Ferreira has a nine-page supply agreement between Orrin Tools Ltd and Vasek Metals Inc., scanned with a received stamp on page one and initials on every page. For the exhibit to her client's declaration she runs Searchable Image with Create a New Document ticked, then checks that the stamp and initials look the same and that a search finds "Vasek". For the redraft her client wants, she runs Editable Text and Images on a second copy, saves it as supply agreement - EDITED COPY.pdf, and reads the price schedule against the scan digit by digit. Only the first file goes anywhere near the court.

On that supply agreement, search the searchable copy for the names of the parties, the date of the contract, and any invoice total in the body. If an exhibit or a pleading page shares the file, search its caption as well. Read the hits on the picture, and keep any rebuilt page in a file of its own.

Other routes to the same searchable image

If all you need is the hidden text layer, the Editor is one of several ways to get it:

• Another desktop editor offers a similar choice between searchable and editable output, described in the post on PDFelement's OCR modes.
• A free command-line tool writes the same kind of layer, covered in the post on an open-source alternative to ABBYY FineReader.
• The post on making a PDF searchable without Acrobat surveys the rest.
• For someone with no editor installed, Schema Law works in the browser on a PDF, PNG or JPEG of up to 20 MB, and its searchable PDF is described on the site as your original file with a hidden text layer. The export formats include Word for the copy you mean to edit.

Whichever route you take, the engine's reading is a draft until a person has checked names, dates and amounts against the page.