(Archived) How do you copy text from recognized text from an image?

[questioner 0]

Hi

I love the text recognition from images, beautiful stuff. I can search my images which is amazing. I tend to think that those text that are found via search should be available for copying to clipboard. It seems like they are highlighted but not available for converting to text?

[engberg 89]

Thanks, we're glad you like it.

Our image processing algorithms produce a fuzzy set of data for search, not a pristine OCR result:

http://www.evernote.com/pub/ensupport/faq#32b9f063-aae3-44f9-9c5f-7d7900b802db

[heather 604]

If you right-click on any particular note and export it to an .enex file, then open that file in an XML viewer, then copy everything between the "Recognition" tags into a word processing program, you can copy the text by doing the following (instructions are for MS Word, but you can adapt them):

Find:

Replace with: ^p

Find:

Replace with: ^p

Find:

Replace with: ^p

Find:

Replace with: ^p

Find:

Replace with: ^p

Find: ^p^p

Replace with: ^p

Trim out all the extraneous code, and bits that you don't want. Enjoy!

[cyberpup 0]

It seems like you should be able to export just the plain text from the file as a .txt file. I realize that a lot of people would use the service to upload photos of text and then export the text & delete the photos in order to keep their free account, but you could charge a "processing fee" for scanning.

My goal is to take an old thesis that I wrote, scan it, and save it as text. I don't want 125 images. Just a text file with the written content.

Is that do-able?

[engberg 89]

If you can make a clean scan of printed text, you may just want to try a standard "Optical Character Recognition" package to convert that to text. Evernote's algorithms are more focused on searching in images of variable quality, so our results may never be as good as off-the-shelf OCR solutions that just handle transcription of high-quality scans of printed text.

[cyberpup 0]

Your service seems to work very well, as when I search for the text I get great accuracy. It is simply an export issue, and your service makes it quicker with an iphone than with OCR software.

[engberg 89]

The problem is that we generate multiple possibilities for every word. This is why our searching works well, but we're not ideal at narrowing down exactly what the "right" word is for each word. So you'd either have a bunch of jibberish for every possibility, or you'd have a lot of wrong guesses.

Standard OCR is (only) good at coming up with one best guess, for high-quality document scans.