OCR: Converting Images to Text with MODI

Joe Schmoe from Kokomo has a scanned image of a 300-page contract. Joe wishes he could search this file for certain rates and terms, but it's an image, not a text file. OCR might be just what the doctor ordered.

Optical character recognition, usually abbreviated to OCR, is the mechanical or electronic translation of images of handwritten, typewritten or printed text (usually captured by a scanner) into machine-editable text. --Wikipedia

One such OCR solution that you may already have available to you is Microsoft Office Document Imaging (MODI), part of the Microsoft Office suite. Let's look at how you can use Ruby and the MODI API to automate the conversion of a scanned document into text.

Installing MODI

MODI might not have been installed when you installed Microsoft Office, so your first step may be to install it from the Office install disks. If installed, you will probably find an icon for "Microsoft Office Document Imaging" located on your Windows Start/Programs menus under "Microsoft Office Tools". If it's not there, go to your Add/Remove Software control panel, select your Microsoft Office installation, and select the option to add features. Then follow the necessary steps, which may vary depending on your version of Windows and Office.

Accessing the MODI API

To begin with, we'll use the win32ole module to create a new instance of the MODI.Document object:

require 'win32ole'
doc = WIN32OLE.new('MODI.Document')

Loading the Image

The next step is to call the document object's Create() method, passing it the name of the .TIF file to load:

doc.Create('C:\images\image1.tif')

NOTE: MODI only works with TIFF files. If your image is in another format (.JPG or .PNG, for example), you can use an image editor (such as Paint.NET or Photoshop) or code library (such as RMagick) to convert it to TIFF format.

Performing the OCR

The OCR() method performs the optical character recognition on the document. The mthod can be called without parameters...

doc.OCR()

...or with any of three optional parameters:

doc.OCR( { 'LangId' => 9, 
           'OCROrientImage' => true, 
           'OCRStraightenImage' => true } )

LangId: An integer representing the language of the document. English = 9, French = 12, German = 7, Italian = 16, Spanish = 10. This value defaults to the user's regional settings.

OCROrientImage: This boolean value specifies whether the OCR engine attempts to determine the orientation (portrait versus landscape) of the page. The value defaults to true.

OCRStraightenImage: This boolean value specifies whether the OCR engine attempts to "deskew" the image to correct minor misalignments. The value defaults to true.

You may find that tweaking these parameters from their default values produces better results, depending on the individual image(s) you're working with.

Getting the Text

Naturally, you'll want to get your hands on the text produced by the OCR process. Each page of the Document is represented by an Image object. The Image object contains a Layout object; and that Layout object's Text property represents the text for that image/page. So the hierarchy looks like this:

Document
    =>Images
        =>Image
            =>Layout
                =>Text

To accrue the entire text, simply iterate over the Document.Images collection and grab the Layout.Text values. For example:

File.open('my_text.txt', 'w') do |f|
    for image in doc.Images
        f.puts("\n" + image.Layout.Text + "\n")
    end
end

Text, But Not Formatting

No OCR process can guarantee 100% accuracy, but I've found that MODI does a pretty good job recognizing text. Results will vary, of course, depending on the quality of the TIFF image. Note, however, that it cannot preserve formatting of tabular data. So while the text in a series of columns may be produced with a high degree of accuracy, that text will probably be produced with one value per line. So...

apple    orange    pear

...comes out as...

apple
orange
pear

Paragraphs of text have, in my experience, been produced with the proper line feeds. Play around with it and see if it meets your needs.

That concludes our show for today. Thanks for tuning in!