Talk:Unsharp masking
From Wikipedia, the free encyclopedia
Contents |
[edit] Apparent Sharpness
Why is the word "apparent" used here? If you take an image and remove the high-frequency components, it doesn't just "appear" soft -- it is soft, since you've filtered out the detail. And correspondingly, if you apply an unsharp mask operator to an image, the result just doesn't "appear" to be sharp, it is very much sharp (or at least sharper), since this operator amplifies whatever high-frequency components were present in the image (proof: fourier transform the operator). mdf 22:56, 10 January 2006 (UTC)
- On the same subject, there's a small tautology in the article that ought to be removed. I edited it, but it still says
- The resulting print appears sharper than one made without the unsharp mask; the apparent accutance is increased.
- The second clause is just a restating of the first; "appearing sharper" is the same thing as "apparent accutance is increased". If someone can explain how the apparent accutance increases (i.e., how the magic trick works), that would be worth noting. +ILike2BeAnonymous 19:19, 15 October 2006 (UTC)
[edit] History
The article mentions the first use in Germany in the 1930's. Curiously, the unsharp mask operation is just a single step of the more general iterative deconvolution algorithm described by P. H. van Cittert in 1931. Is this a coincidence? mdf 22:56, 10 January 2006 (UTC)
[edit] Unsharp Mask Creates Illusion?
The intro reads: 'creating the illusion that the resulting image is sharper than the original.'. I was wondering if this is actually correct. Shouldn't it be 'resulting in a sharper image than the original but without increase of information'?
This clearly depends on the definition of 'sharper', which is in my opinion 'a more clear distinction between objects', which is exactly what the unsharpen mask does. It does not mean 'more information' I think. 71.206.215.239 01:16, 13 February 2007 (UTC)
[edit] How about an example?
I think this article would be improved by adding at least before and after images, maybe with intermediate steps.--Jwwalker 03:19, 1 November 2007 (UTC)
[edit] Frequency response
As a series of convolutions, isn't an unsharp mask a linear filter? As such, can't we describe it in the frequency domain by a transfer function? If so, is it a bandpass filter? Since it is usually constructed from Gaussian blurs, its transfer function should be nonzero almost everywhere... —Ben FrantzDale (talk) 22:50, 21 January 2008 (UTC)
- Duah... the FFT of a Gaussian is a Gaussian, so the transfer function should also be a Difference of Gaussians. Sound right? —Ben FrantzDale (talk) 01:36, 22 January 2008 (UTC)
-
- Well, no. Or yes, sort of. The filters used are not necessarily linear (that is, they can use the "threshold" or other nonlinear mechanisms). But a linear analysis is certainly a useful part of it. They're also not usually Gaussian, because a filter that falls less rapidly in the frequency domain will make a better looking result. The article says Gaussian blur, and that may be true in some cases, but it's unsourced and I wouldn't trust it. And overall it's not a series of convolutions, it's a convolution and a difference and an add, but still there's an equivalent linear filter; it's usually applied in a nonlinear (gamma compressed) domain. Dicklyon (talk) 01:47, 22 January 2008 (UTC)
[edit] Dubious
I just added the "dubious" tag. Here's the thing: As a linear filter with, unsharp mask doesn't add any new information to an image. It makes the image look sharper by enhancing the high-frequency components, but it doesn't add any missing data. Deconvolution in its simplest form can't either, although it is often done in a more sophisticated way than just an edge-enhancing algorithm. I may just trim that section... —Ben FrantzDale (talk) 01:41, 22 January 2008 (UTC)
- I think what that section says seems true, as far as it goes. It would be better if it were sourced. What are thinking is worth removing? Why not add a fact tag if you're not sure? Dicklyon (talk) 01:47, 22 January 2008 (UTC)
[edit] Sharped ? Or Sharpened ?
Does Wikipedia's version of the English language permit nouns to be 'verbed' at random ?)
I would let 'bolded' text pass, though - 'emboldened' is an ugly word, and tends to mean 'encouraged' ...
--195.137.93.171 (talk) 01:25, 7 March 2008 (UTC)
[edit] A schematic illustration would help
I'm having trouble visualizing the analog process with the glass plates. It would be neat to have a side-view illustration showing the light from the enlarger passing through the original negative and the blurred positive, and the resulting brightness level striking the final print.
In my head, it feels like all this should do is make dark areas lighter because the dark areas of the positive are blocking the light which (because enlarging is working with a negative) would darken regions of the print. I'm failing to visualize the process, so if someone who understands it could sketch out a diagram rather than just before/after pictures (which any of us can see in our paint program, thank you), it would be very helpful. Myself248 (talk) —Preceding comment was added at 01:12, 13 May 2008 (UTC)
- Presumably one using this process would increase the exposure to compensate for this problem. If you had a low-frequency sine wave, it would look almost exactly like a Gaussian-blurred version of itself, so if you make a positive of it and project its negative on it, you'll get a uniform brightness out the other side. If you have a high-frequency sine wave and blur it, you get medium gray. Project the original negative through that medium gray and you just have to compensate for exposure but will still get the same high-frequency pattern. Does that help? —Ben FrantzDale (talk) 02:06, 13 May 2008 (UTC)
