Unbluring a photo in photoshop

I think people have used the channels in photoshop, and accomplished this feat, but it's not reliable.

Hi George,

Data that is not in the original file is simply not there. You could try the Unsharp Mask filter and play with different settings for Amounts and Radius but you'll only get a smart increase in contrast to simulate a sharper image, details that were not visible in the original because of the blurring will not show up out of nowhere. Perhaps a small amount of Unsharp can help make the images LOOK less blurry but I don't think you can make them sharp as if they were taken right on focus.

Good Vibes!

Lobo

LObo is right. You can't "focus" an out of focus image with any kind of tool. The blurring means loss of definition, loss of information. You can't create something out of nothing.

I use a plug-in called Focus Magic to do this. It can correct two types of blur:

* Out-of-focus blur
* Motion blur

It really is magic, but don't take my word for it -- there is a free trial so you can see what it can do. It's at

http://www.focusmagic.com/

By the way, Jason210, the reason that Focus Magic can work is that pixels surrounding a blurred pixel contain information about what the correct value of the pixel should be. So it isn't really true to say that you can't unblur a photo, as not all the information is lost -- there's a remarkable amount in the neigboring pixels. You can use convolution algorithms to extract the information from surrounding pixels to remove the blur.

The same concept is present in Bayesian statistics, where multiple samples of the same value, even if they are in error, result in a more accurate estimate of the true value.

I was thinking about it. If the dispersion pattern through which the image got blurred can be calculated then the pixels could be "undispersed" by reversing the pattern. In order to achieve this, however, the exact amount of defocusing or degree of movement would need to be recorded. I believe that is the idea behind Focus Magic. Nevertheless, since the actual dispersion can't be calculated (since there's no way to record the amount of defocus in an image) the program does a "best guess" job to correct the image. The results are not perfect but they can fool the eye sometimes. I'd say give it a shot.

Another thing that was going around in the back of my head was using different software not to actually sharpen the image but to give the impression that it has been sharpened. Genuine Fractals is a Photoshop Filter that uses fractals to allow a user to upsize an image without noticeable loss of quality. So, the crazy idea was to resize the image to 600% using Genuine Fractals, and then shrink it back to its regular size. I haven't tried it but I think the result would be an image that will look much sharper than the original.

Problem is that, as with Focus Magic, Genuine Fractals is not free and is not cheap either. But the free trial lets the user run it 20 times as a test drive. Downloadable from:

http://www.altamira-group.com/products/gf/

Good Vibes!

Lobo

Well it depends on how it's blurred - lobo has it right.

I remember seeing a geometrical paper showing why information is lost in a blurred image, although if, as Lobo says, the dispersion pattern is known, then some of the information might be restored.

Thanks for the link.

Lobo,

It works like this:

For out of focus images, what should be a pin-point of light becomes a circle -- the "blur circle". Thus the shape of the dispersion is known. The program can estimate the radius of the blur circle automatically, but you get best results when you try different values and see what they look like in the preview window. Furthermore, Bayesian statistics shows that it doesn't matter if the adjacent samples are in error -- you can still use them to accurately estimate the true value.

For motion blur, you get a line. In the program, you have to tell it the angle of motion and thus the program knows the shape of the dispersion. It can then correct for motion. Same comment re: adjacent samples being in error.

What you need to remember is that not all the information is lost. While convolution algorithms cannot bring a badly out-of-focus image to complete sharpness, they can do an extremely good job. And on slightly soft images (such as the softness caused by the anti-aliasing filter in many digital cameras), they can do a stellar job.

I would suggest that anyone interested in this download Focus Magic and try it for themselves as this will help you understand my explanation.

mjlodge,

I do understand the theory behind it. However, observing the samples at their website it's easy to see that many details have been "restored" wrong. In the image of the couple you can see a strange dark line running along the lower edge of the guy's chin that makes it look like he's got a "double chin" but out of place. I don't see any suggestion of a double chin in the original blurred image. In the old car image, there are small artifacts on the cabin walls that suggest the kind of artifacts Genuine Fractals generates to simulate sharpness. The image of the red car has even more of these fractal artifacts pretty much all over it, they're easier to notice because there are more high contrast edges (like the car's grille, i.e. which looks anything but the smooth surface it should be.) The image of the beige car doesn't show these artifacts but looks very grainy instead, which suggest a differnt algorithm used by the program, perhaps? The image of the safari helmet shows a sharp edge on the right side but the blurred edge on the left side hasn't changed with respect to the original blurred version.

I'm not saying it's a bad program, though. But I would not call its results stellar, either. It's an improvement and a good one for non-critical jobs but I would not use it for professional work just yet.

Good Vibes!

Lobo

Er... thanks for the points, but I don't deserve them. You should ask for this question to be re-opened, then assign the points to whoever is worthy. I think you should share the between lobo and mjlodge.

Incidently mjlodge, I tried Focus Magic, but I couldn't really see the difference between this an the unsharp mask tool. I only had 10 goes with the demo, which I wasted on trial & error attempts at different pixel radii.