Any free software to get hi-res image from lo-res video ?

Bean

I have a camera that is VGA resolution.
I've heard that there is software that can take a number of frames and create a much higher resolution still image from it.
I've done some searching, but I can't find any programs that do this.

Is this not really possible ? Or is there just no free or inexpensive software that can do it ?

Thanks for you help,
Bean

Phil Pilgrim (PhiPi)

What you're talking about should be possible, albeit certainly not easy nor fully automated. What would make it work is when a subject moves by a fraction of a pixel, so that different parts of the subject overlap within the individual pixels. From that info subpixel data could be extracted from multiple shifted images.

In reality, though, strictly linear motion between frames would be rare, so tracking minute parts of the subject from pixel to pixel won't be easy. Still, though, software that could do this would be invaluable for identifying miscreants from typical poor-quality security camera footage.

-Phil

Beau Schwabe

Do a search for "Unlimited Depth of Field"... The idea is that you take multiple photos of the same scene. What tends to happen is that any random noise in the image remains at the "photographic noise floor" while the actual "signal" or content within the picture reinforces which increases your signal to noise ratio and you have an overall better quality picture. For every N samples or pictures you take, the SNR increases by the square root of N. This goes right along with the recent article I wrote for Nuts and Volts magazine, virtually the exact same concept.

Here is a reference to "Unlimited Depth of Field":
http://www.techhive.com/article/131096/article.html

...and to my Nuts and volts article:
http://nutsvolts.texterity.com/nutsvolts/201603/?folio=34&pg=34#pg34

Dave Hein

The easiest way to create a higher res image is to pan the camera left and right, and up and down, and then use a stitching program to splice all of the images together. You could almost double the resolution just by stitching together four pictures that cover the four quadrants of the hi-res image. There just has to be a small overlap among the four images so that the stitching program knows how they connect to each other.

In theory, you could do what Phil is saying, but you would need many images and the software would have to invert the point-spread function of the camera. Even with many images the result probably won't be very good.

"Unlimited Depth of Field" is different that SNR improvement. Unlimited Depth of Field splices together the "in-focus" portions of many pictures that have been taken at different focus positions. SNR improvement takes advantage of the uncorrelated nature of noise versus the correlated nature of the image itself. However, neither method increases the resolution of an image.

Bean

Focus isn't the problem, I'm looking to get a higher resolution image.
What I'm looking for is to be able to make out a License plate of a car that I follow.
The VGA resolution is just not quite clear enough to make it out.
Bean

kwinn

Bean wrote: »

Focus isn't the problem, I'm looking to get a higher resolution image.
What I'm looking for is to be able to make out a License plate of a car that I follow.
The VGA resolution is just not quite clear enough to make it out.
Bean

There is software out there to sharpen images, but I have no idea of who has it or what the cost would be any longer. Probably simpler and less expensive to upgrade to a higher resolution camera when all is said and done. A possible alternative to that might be a lens with a narrower field of view or a zoom lens.

mindrobots

Gimp (and most any other photo editor) has blur/sharpen tools that may help you with sharpening a digital image. Basically, it makes the differences between pixels a matter of black/white and gets rid of shades of gray(or color). You can sharpen an image to the point where it is horrendous to look at but may pull out the details you need.

(*sharpening is a common practice in post-processing a digital image. You often hit the eyes of a portrait or some details of other images with a bit of sharpening or blurring in some areas)

You may also want to try an HDR tool. There are some free ones but I can't find one off the top of my Google. The HDR overlays multiple frames to increase the dynamic range of the image. (Much as Beau describes above) This process can lead to increased detail but will also be tripped up by any movement in a frame, not knowing if a pixel changed due to changes in lighting or movement pixels as Phil discussed. Most software has tolerance values that can be adjusted to play with how it impacts the image.

The sharpening could most likely be applied in real time but the HDR would be a post-capture process.

This would be fun stuff to play with on a Raspi with it's camera (and its IR camera). If you are particularly interested in license plate capture, you may want to look at how license plates respond to natural (or unnatural) IR most states have a reflective coating of some sort. For $85 or so you can get a Raspberry Pi and both camera modules and play with image processing on the Raspi. I better shut up or I will distract myself!

Phil Pilgrim (PhiPi)

Bean, can you provide a sample image that shows just how bad the current resolution is on a license plate? There may be other ways to skin this cat.

-Phil

tonyp12

https://articles.forensicfocus.com/2014/10/08/can-you-get-that-license-plate/
http://www.focusmagic.com/forensics-examples.htm
http://forum.videohelp.com/threads/334150-Enhancing-a-still-from-video

John_sky

You can find and use portable photoshop. That is free and have this feature.
I usually take a snapshot and edit it in Photoshop.

erco

Youtube has several enhancement features.

https://support.google.com/youtube/answer/1388383?hl=en

jones

If individual frames don't have enough resolution to allow the plate to be read, it might be possible to combine frames to get enough information to resolve the numbers. There's a program called Registax that's used principally by amateur astronomers for creating very high resolution planetary images. What they typically do is shoot hundreds of frames (or more), let the program sort which frames have the best "seeing" (e.g. minimal atmospheric distortion) and then the best are aligned and combined. In this case the sorting step might not be necessary, but the alignment certainly will be. As Phil said, it's the sub-pixel sampling that will increase the resolution of the combined image. There are other tools that are included, like unsharp masking, etc. that can be applied to the combined frame. It may or may not work, but it might be worth a try. The program is free and it's designed for sequences of video frames.

http://www.astronomie.be/registax/

msrobots

jones wrote: »

If individual frames don't have enough resolution to allow the plate to be read, it might be possible to combine frames to get enough information to resolve the numbers. There's a program called Registax that's used principally by amateur astronomers for creating very high resolution planetary images. What they typically do is shoot hundreds of frames (or more), let the program sort which frames have the best "seeing" (e.g. minimal atmospheric distortion) and then the best are aligned and combined. In this case the sorting step might not be necessary, but the alignment certainly will be. As Phil said, it's the sub-pixel sampling that will increase the resolution of the combined image. There are other tools that are included, like unsharp masking, etc. that can be applied to the combined frame. It may or may not work, but it might be worth a try. The program is free and it's designed for sequences of video frames.

http://www.astronomie.be/registax/

+1 for that. A friend of mine does amazing images with his home build refractor telescope and combining images.

Enjoy!

Mike

potatohead

I gave that software a quick run today. Took an image and applied various degradations, distortions, and noise. It can take discrete frames and make an AVI, or just work from same.

Impressive! Great find. I'm glad it was linked here.

I think I'm going to take some of the company CCTV footage and see what can be resolved.

Bean

Thanks Jones. I'd give it a try.
I kind of have my doubts though. The plates are only maybe 20x10 pixels or so.

Bean

jones

Bean wrote: »

Thanks Jones. I'd give it a try.
I kind of have my doubts though. The plates are only maybe 20x10 pixels or so.
Bean

I'm not that optimistic either, but I thought it might be worth a try, especially since it's free and requires no hardware/firmware changes. The simplest solution would be a longer lens, but that would cost you field of view and would make focus more of an issue since increasing focal length reduces depth of field.

GordonMcComb

Bean wrote: »

Thanks Jones. I'd give it a try.
I kind of have my doubts though. The plates are only maybe 20x10 pixels or so.

In that case, your best bet is indeed some of the apps for amateur astrophotography. Some of them can extrapolate an object or texture, based on just a few pixels. There are some generalized techniques here:

http://www.skyandtelescope.com/astronomy-resources/astrophotography-tips/how-to-process-planetary-images/

There are just barely enough pixels in a 20x10 grid to form unique character shapes, and that's assuming no camera shake and a straight-on view.

If you're doing this for a commercial or non-commercial project, the best method is to use vision analysis to "find" the license plate in the wider field, and use quality optical zooming to take a succession of multiple frames, preferably at frame rates exceeding 240 fps.