One of the things I'm having trouble understanding here is the context or contexts for the OCR processing being discussed:

1. At the time of scanning an actual print document on a flatbed or feeder scanner.
2. In an existing document, such as an image scanned PDF or an actual image itself that is known to be an image of a page of text.
3. "On the fly" in something that's an image being presented in, say, a web browser that one would like to have read.

Virtually any relatively recent, and by that I do not mean new, scanner or multi-function machine will come with an OCR engine as part of its manufacturers software suite.  Canon, HP, Kodak, and other multi-function machines I've worked with were all able to scan with OCR as part of the scan and, as it happened, their scanner control software was also accessible by screen reader.  I've been using Canon's multi-functions for years now, and even the cheapest one (read: cost about $30 ten years ago) had a software suite included with it that featured OCR as an integral part of scanning if one identified what one was scanning as a document.

I've repeatedly mentioned that Tracker Software makes two excellent free pieces of software that, while not 100% accessible, are 100% accessible for doing OCR processing on image PDFs with a remarkable level of accuracy and that also feature free language packs if you're not scanning documents in English.  See either PDF-XChange Viewer or PDF-XChange Editor.

The "on the fly" instance is one where I don't know of anything off the top of my head.

--
Brian

 He discloses the workings of a mind to which incoherence lends an illusion of profundity.

         ~ T. De Vere White