Hey, that’s my car!
I sent a copy of the orginal to the email in your profile. I tried to see if I could post it to flickr, but it appears I would need to upgrade my account.
Try uploading it to http://www.imageshack.us I don’t think they resize photos
OK, I think this is the original:
Here’s a crop of just the car and person (no change to resolution and saved as png so as not to introduce more loss in compression).
I was trying deconvolution using Image Analyzer, but there are other advanced methods possible - for example, as detailed on this page:
Digital Image Processing (I don’t know how to implement the methods described there though)
-The level of detail recovery in those example photos looks like sci-fi, but really, it isn’t - because the original image is defocused in a consistent manner.
Im voting this is impossible. If you play with the image and look at the plate you’ll see that the bottom 2/3rds of the plate is reflecting sunlight, thus causing the cameras sensor to overload. Brightness/contrast dont help and playing with the RGB levels just shows the distortion even more.
The best I could get was this, which kinda sorta maybe looks like the first character is a 5 or an S. Which might help the cops as they can do database searches with model/color plus whatever plate numbers they think they have.
Okay, I maintain the idea that it begins “664” and now think it ends in a “Z”.
But I don’t know what the regular order of letters and numbers should be, or if there even is a consistency for that. According to this page, it ought to be either three letters then three numbers, or one letter, two numbers, three letters.
In which case, 664##Z could also be GGA##2 or G64##Z
In fact, if I squint I see (with the gap filled by a graphic): 664 TZ
Okay, I also ran my own best contrast/color combo via the FocusFixerV2 plugin and got this:
Im definately seeing a 5 for the first character. Everything else is just a distorted mess.
I just got an email from the Oldsmobile dealership. The car is an Oldsmobile eighty eight from '90-'95. I am compiling the images and suggestions that you are all coming up with and plan to give it to the police officer that seems most interested in tracking these guys down. I am hoping that they will run some combinations of possible plates and car info.
I know I sound a bit repetitive, but I really appreciate all of you taking a look at this and trying to help. Thanks again.
I wanna say it starts with a 6 and ends with a 7, but the 7 could be misleading, the top part of the 7 is a shadow so I’m going by the angle of the vertical part of the 7, but it could be a 2 or a Z. It would be nice to have Florida plates with all the letters for comparison.
Maybe it says “STEALZ”
http://img32.imageshack.us/img32/4941/snap2his.jpg
http://img269.imageshack.us/img269/8317/snap4x.jpg
I think the info IS there. I’ve looked carefully at the original image.
The plate area is closed to overexposed, but I dont think it is. Most of the pixels in the tag area look to have at least slightly different intensity levels, so very little appears to be totally burned out.
We also know exactly what color the letters should be, a green/green with a hint of blue. We also know what color the orange/oranges in the center are. You should be able to play with the color channels and contrast/intensity ratios to help bring out the details.
More importantly, is next.
There definitely ARE enough pixels in the tag area to easily read the number if it was crisp. The area is almost readable as is.
The lack of resolution is due to either focus blur, lack of resolution of the lens (high f stop), or camera movement. It is NOT due to a lack of number of pixels in the area.
The first 3 of those things are somewhat correctable. Let me be more specific. Look at the bottom edge and right edge of what should the edges of the tag. We know those should be a sharp edge, one pixel or two pixels wide at max. Instead they are 5 and perhaps significantly more pixels wide.
You should be able to run a 2D deconvolution filter horizontaly using/recovering the right edge. You should be able to do the same vertically using the bottom edge.
Keep in mind that the image is basically cropped by a factor of 2.8 and then enlarged by the same factor, as indicated by the “digital zoom” field in the EXIF data. At least if my understanding of digital zoom is correct. I thought they just cropped without upsizing again though. That is a loss of information that is very likely to put success out of reach, and isn’t a convolution like the optical aberrations of the lens.
I am playing around with some deconvolution tools anyway, but I need some way to estimate the point spread function of the convolution in order to reverse it. That is lens and sensor dependent though. Hm, in order to get that some sample photos of point light sources would help, at the same distance and camera settings. Maybe that could also be found by using some other part of the image as a reference, but I have no experience with this kind of stuff.
Next time, your friend should not use zoom, but try and hold the camera still and focus for a few seconds so the image is sharp.
What stops you from printing the picture and distributing it on your block? Or to post it on Craigslist? Or even approach the local press? The picture is good enough for that. You say the burglars have been there a couple times, so there is a likelyhood that others have seen them too. You might get more confirmations, more and better pictures, maybe someone even took the plate down. Or your neighbours will be on greater alert and the robbers will be caught, if the robberies continue. Approaching the press and have them run the story (I would, if I was press, it makes an interesting copy especially with this pic) might shame the police in doing somthing.
Sorry but I’m not getting any better results. I do see the 5 as the first number though or maybe a 6
Okay, so the pixels I am looking are NOT the pixels on the sensor, they are “fake” extra pixels in the data so to speak. When I zoomed in I thought there seemed to be alot of them :smack:
Can’t you estimate the point spread function by looking at the right edge and bottom edge of the tag? You know its a sharp edge right in real life.
I’d play around with making those edges look as sharp as possible without making them look weird, and then see what the letters look like.
Oh, you could get the person that took the pic to take another under similar conditions. Just tell them what you need them to take a pic of to estimate the point spread function.
Another distinguishing feature on the car is the little “blind spot” mirror that has been affixed to the driver’s side mirror. That might help identify the car if someone happened to encounter it.
I don’t think that’s a mirror, a blind spot mirror would be stuck on at the far left side of the regular one.
That’s a reflection.
I am trying not to sound horrible here, but really–I hope there is no next time that my 17 year old neighbor witnesses break and entry crimes or any other crime. I applaud the action that she did take and I cannot imagine going to her now and saying that she should have done things differently including holding the camera steady, while on the phone with 911.
Additionally, we have distributed the photo throughout the neighborhood, called a meeting of the historical preservation society to ensure that we are getting the most coverage. We have considered approaching the press, however at this point it may be considered overstepping and antagonizing the police force. This is an area that has a fairly sharp distinction between relatively wealthy people and those who are destitute. There are those of us who love these old houses have put a bunch of money, time, love, etc, into restoring the places. This means that the squatters, hookers and thugs are angry–and we are asking the police to pay attention to something that they haven’t had to–people who care. People who want to restore this area and love living here. The shift in paradigm will take time. I understand that we have taken a risk by living in this area–I still love it, even after the break in.
So, after a little too much TLDR–Thank you all for your help in trying to sort out this plate. I have some more information to give to the police. It will be passed on.
The problem is not camera shake or incorrect focus but the resolution limit of the sensor/lens combination. I think the picture is as good as it gets for that combination of camera and distance. If that was actually the problem, it would be possible to use some techniques mentioned to retrieve the plate.
This is about what a 100% crop of the original image would look like without the camera’s upsizing, which seems to be what one might expect from that sort of camera.
