Enhance this image to get the license plate

[thestolz]

Hi
Someone hit my car and I was able to get the video footage. Here is a capture, is there any way to get the plate number from this image? Any help would be greatly appreciated

[KarMa]

Can you post the video clip here. Don't re-encode it, and don't upload it to youtube, just post the original file. Apologies if I sound harsh, as these threads get posted quite a bit and people tend to do everything but post the original video.

[thestolz]

Sorry I will post it momentarily.

[thestolz]

Here is the video.

[thestolz]

http://files.videohelp.com/u/260647/ch00000000000001-160629-064040-065100-10p00001000000400.264

[KarMa]

I took your raw .264 video and muxed and cut it up for an easier to view mkv, which does not alter the quality of the video. Anyway the source video quality is pretty low, and the video sensor used in the camera seems to be also on the low end also. Which means that fine details, like license plate numbers, are hopelessly lost. The license plate numbers need to be pretty close to legible as is, before any "enhance[ments]" can help.

[Leo E. Backman]

thestolz,

Hi,

Is the case of unreadable license plate still active?

I've downloaded the video to see if the plate recognition is doable. The surveillance camera image resolution is rather limited. Based on viewing single frames, the plate looks pretty illegible. However, the time in which the car is closing in and bumping into yours, the camera has recorded maybe 20 or so frames -- with relatively little frame-by-frame movement

Now, I'm thinking of separating these 20 plus frames and stacking them into one image, while correcting the plate's incremental movement in the frames. The idea is to make the plate appear in one place in all images of the stack. A slight change of the plate size should also applied for since the car first gets closer then stops and then starts getting farther from the camera.

As a result of properly summing these frames together, the plate's fuzzy features (letters and numbers) will be reinforced, maybe rendering the features legible. Of course, the operative word here is: may. But let's give it a try.

cheers,

LeoB

[thestolz]

The case is still active and any help would be appreciated!