Extrapolate letters from blurry letters (e.g. license plate)?

Question

My friend has a blurred image of a thief's license plate. Is it possible to run an algorithm on these pixels to determine the most likely characters that the pixels represent?

(The fact that this is a license plate does not matter, the solution should work on the principle of any photographed text that is difficult to decipher.)

Please help me find my dog!

Update: my friend sent my two still images, they are very poor. Unfortunately, he does not have license number frames.

Image 1: Image of the van. I am trying to decipher the text above the word "rent".

Image 2: This is the same van, presumably the same text. It is located in the upper right corner of the image.

Answer 1

It depends on why / how it is blurring. There are a few things you could try: one would be a simple sharpening with a blurry mask. Another thing that I have found surprisingly effective from time to time is to simply invert the colors in the photo - sometimes things that are really hard to read usually just appear immediately after inverting.

However, for a one-time task like this, you probably want to use existing tools (like Photoshop or Gimp) instead of writing new code. It will take a long time and require a lot of effort to match what they already expected to be usable.

Answer 2

Play with Photoshop. Try different sharpening filters, with different strengths and different orders. Also play with posterization. Often return to the original image. Look for what works. Use your eyes. If you do not see the answer (after applying the filters), OCR will probably not be either.

Answer 3

Motion blur can be deleted because all the information is still in the photo. But in this case, I'm not sure if any form of image processing will help.

I apologize if you have already tried this, but looked through the "rental" section of the phone book to find out if you can find a company with a similar logo? My guess is that this is a rental car or rental business van, not a car rental agency.

Answer 4

I would recommend sharpening and then a Sobel filter to find the edges and then perform OCR on it.

works:

http://en.wikipedia.org/wiki/Sobel_operator

http://www.bythom.com/sharpening.htm

Answer 5

Theoretically, this is possible under ideal conditions. But this requires that you know the conversion from the original to the blurry image.

Image compression, camera non-linearity, limited resolution, and noise can interfere. If you're lucky, then he will make a standard sharpening filter for Photoshop.