Image enhancement, any good freeware?

[Ulysees]

I've got a terrible picture from a mobile phone's camera. Can someone recommend good software for image enhancement?

For starters, what is the best method for resampling that is available in freeware?

 

[robphy]

neatimage (has a free mode)
paint.net
gimp
imagemagick

 

[Ulysees]

Thank you robphy. Have you put them in order of capability?

 

[CaptainQuasar]

Depending upon what you're trying to do, there are different answers here I think. If you're trying to enhance the image so as to see details you can't currently, that's one thing. But if you're trying to find a way to enlarge the image without pixelation, a vectorizer like http://vectormagic.com/" can help depending on what the photograph looks like.

⚛​

 

[Ulysees]

Depending upon what you're trying to do, there are different answers here I think. If you're trying to enhance the image so as to see details you can't currently, that's one thing.

This is what I want. The only way I knew was with paintshop's "pixel-resize" and "Gaussian blur". Not very good. What's a proper way to see detail? Saw an article about fractal-based enhancement once in Scientific American, I wonder if it's available for free and what opinion people have of it.

 

[binzing]

I've got gimp, just got it recently. Seems OK, I haven't had much time to mess with it. I'd recommend Google Picassa for cataloging and basic editing.

 

[CaptainQuasar]

I think people may be misunderstanding you, Ulysses. To my knowledge the software people have listed is all basically just image-editing software.

I think you can sometimes get some mileage out of adjusting the contrast and color palette of an image. But the kind of stuff that does real hardcore image enhancement - like being able to pick the license plate digits out of a blurry security camera video - I do not think is the sort of thing you can do for free, or even very cheaply, unless you yourself have a good understanding of the math and science involved (which I do not, personally.)

⚛​

 

[Moonbear]

I think the blunt answer is that if the original image doesn't have the detail, unless you're drawing it in, you can't get more detail than you started with.

 

[CaptainQuasar]

Oh, wow, totally ignore what I said above. Sorry to fake everyone out, it's not so expensive after all at this point. This company provides a free trial of their software Focus Magic, it looks like:

http://www.focusmagic.com/exampleforensics.htm

And that was just the first result of a Google search for http://www.google.com/search?q=image+enhancement+license+plate+number". So there are probably others out there too.

Focus Magic looks pretty good, similar to the professional forensics stuff I've seen in years past. And it's only $45 even if you want to pay for the full version. Check this out:

http://www.focusmagic.com/images/forensicscar2b.jpg ➧http://www.focusmagic.com/images/forensicscar2a.jpg

⚛​

[/URL]

 

[Ulysees]

That's way cool!

I wonder if it cheats though, by having predefined patterns of plate characters and forcing them onto the image in the right transformation, as a probabilistic choice of 1 out of N patterns, choosing the one with the highest probability in each case.

 

[CaptainQuasar]

It might. But that particular example was an out-of-focus picture, so it seems feasible to me that it could be accomplished by something like reverse raytracing. When I've read articles about this technology they've mentioned “nonlinear differential equations”, but without describing the technique in detail.

⚛​

 

[robphy]

Thank you robphy. Have you put them in order of capability?

It's in the order that I thought of them.
They have different strengths... and may or may not be useful for your particular problem, which you haven't clearly specified.

NeatImage is for digital camera noise reduction.

paint.net, gimp are photoshop-like image processing programs
imagemagick is a command-line image processing program

paint.net, gimp, imagemagick support plugins and scripts that can help do intricate operations and automate the process

I'll add in ImageJ, which is useful for [say] medical image processing. There are lots of scripts for ImageJ.

There are free vectorizers like autotrace.

Everything listed is free (although NeatImage's free mode has some limitations).

...but as it's been said... it really depends on what you need it to do.

 

[Ulysees]

They have different strengths... and may or may not be useful for your particular problem, which you haven't clearly specified.

I have not specified a particular problem at all. This is NOT about just reading plate-numbers, or just reducing compression artifacts, or just interpolating based on fractals, or just reducing camera noise, or just whatever. I want the best general-purpose freeware there is for image enhancement, and appreciate your evaluation of each of the programs mentioned, strengths and weaknesses, thanks a lot.

 

[_Mayday_]

I think the blunt answer is that if the original image doesn't have the detail, unless you're drawing it in, you can't get more detail than you started with.

This is true, you can play around with contrast and brightness but you cannot improve the quality. If it is blured or is out of focus there is little you can do...

 

[Ulysees]

Much of image processing is about taking advantage of known properties of images. Images are not arbitrary signals. So interpolation can be much smarter than just a bilinear or a bicubic.

To get the idea, it's like this: If you know a signal is from a piano being played, but it is low-pass filtered to 500 Hz and noisy, then you can derive the rest of the spectral content and almost eliminate noise by taking advantages of known spectral properties of notes on the piano.

 

[Ulysees]

Provided there are no notes with a first harmonic higher than 500 Hz.

 

[CaptainQuasar]

Looking at those Focus Magic images I'm noticing - see how there are chromatic aberrations in the enhanced image, the blueish and yellowish patches? I wonder what that says about the kind of processing they're doing.

Ulysses, it might be worthwhile to go search for a few users who are obviously optics experts and PM them with a quick invitation to join this thread. (Just don't spam a gazillion people, that would be bad form.)

(Oops… on preview, maybe you're an optics expert yourself.)

⚛​

 

[mgb_phys]

I wonder if it cheats though, by having predefined patterns of plate characters and forcing them onto the image in the right transformation,

That's how most of the pattern recognition ones work, you use either Maximum entropy or Baysian stats to get the most likey image that blurrs back to the source data.
It was originally developed for radio astronomy where you are trying to reconstruct an image from the tranform of a sparse and noise U-V plane.

Fixing out of focus and camera motion are relatively simple because you kno at transform to apply - and can ork very well if you have high enough signal to noise in the image.

 

[Ulysees]

it might be worthwhile to go search for a few users who are obviously optics experts and PM them with a quick invitation to join this thread.

I think there is already a good selection of software to get started with. Especially Focus Magic.

 

[binzing]

I'd guess that it (Focus Magic) works by shading darker areas (letters) and lightning lighter areas (empty space)

 

[Ulysees]

It was originally developed for radio astronomy where you are trying to reconstruct an image from the tranform of a sparse and noise U-V plane.

With jargon like "a U-V plane" it gets a little cryptic but are you suggesting that astronomers force stars to look like circular disks?