How to OCR engrave text?

Question

I have this image

How to do it? I know this is very difficult, but I would really appreciate any help.

Answer 1

If you have time to develop the discovery yourself, I would do something like this:

• Get 1000 images or so and either OCR them yourself, or let people on Amazon Mechanical Turk do it for you, it will be practically nothing. Now you have something to set up your algorithm and measure how well you do it.
• As Ryan wrote, play with standard image filters, contrast, color, gauss, etc. manually or with something like http://www.roborealm.com/ . See if you can find a combination that makes the text really stand out.
• Try the libraries again
• If libs still does not work, try using your knowledge of the painting to separate it into separate digits. You know how many digits there should be and how many pixels there will be. Use edge detection or something like that (maybe standard OCR extraction along with clustering will give you each digit as a cluster?) To find the digits and separate them separately.
• Perform standard extraction of the OCR function (not too creative - use existing libraries or at least read that most often and simply) on each digit, and pass these functions along with the answer that you received under 1) to a neural network or SVM .
• Improve your feature set until your computer starts up.

Since you only have ten digits that are reasonably consistent between the images, this should work.

Answer 2

I would suggest you 2 libraries:

• Tesseract
• Emgu CV - comes with many examples looking for a license plate as a suitable place to start.

Answer 3

Try to reproduce the contrast and gamma on the image. All you need is a solid character outline for most libraries. Depending on your SLA performance, you can run different contrast / gamma scenarios and let the OCR software take multiple shots. Take a set of results and see if there are any consistencies. This can give you a fairly accurate result in the long run.

Answer 4

ML (neutral network) for numbers is usually accurate already with minimal training and easy to use. The order can be handled by OCR-ing with a "moving window" or how you cut out a tiny fraction of the width. The output can be 1160060060 ?? 1??? for the first character, and you select the most frequently occurring (0) iterations over the width of the image. Perhaps teach your neutral net to also recognize the space between the figure and your good. Clear ten-fold otherwise requires automatic automatic cropping. But overall, it is very similar to the task that you will receive in AI beginner training courses.