Image Manipulation: Convert embossed letters to normal letters
I'm searching for a way to convert embossed letters e.g. like in the picture of the credit card below to normal ones like you see them in a book. I don't need specific code here, just ideas which image manipulation steps could lead to the desired result.
One thing you do know is that it's a Credit Card number -- this will help validate the result at the end. You could use this to fine tune the image processing -- feedback the result and use optimization techniques (like hill-climbing or genetic algorithms) to change the parameters of the image processing.
For example, do the image processing with a parameter set (p1, p2, p3, p4) -- then try OCR and you get 6 numeric digits. Then alter the parameter set -- say this time you get 8 numeric digits -- use hill climbing to keep altering the set.
For image processing
- Probably want to start by cropping to number regions if you know them.
- Clear pixels that aren't in the embossed color range to take out as much information as possible (the tolerance of this is one of the parameters in the parameter set I described) -- probably do this by setting it to black because the part of the numbers that's easy to see is light
- Target the light color of the emboss -- try to extract it, then invert because OCR finds Black on White
- Connect close sections
- OCR and feed results into parameter choices in step 2, 3, 4 (and maybe 1 if you are doing automatic segmentation)