(29/11/2010 07:09 PM)Assassinator Wrote:  Any other interpolator, which produces a curve smoother than a staircase (ie. everything else), should be less effected by "special" points than the nearest neighbor method, thus the artefacting would get "smoothed away" to some degree compared to the original.

They get smoothed, but enlarged at the same time.

Take an example: let's say wee have a 100x100 white image, with a single black vertical line, 1 pixel wide drawn down somewhere.
If wee downscale it to 25x25 using nearest neighbor, the line is either going to disappear (become 0 pixels wide) or stay the same width, depending on where the line is drawn.  If this position is random, according to probability, the line will most likely disappear in this case.
You've mentioned an average, which works in this case because nearest neighbor is non-discriminatory, as you've said.

Other scalers, however, can be discriminatory.
If I did the above downscale, but use, say, a Lanczos scaler, it still has the problem of making 1px wide line "smaller".  You can't go below 1px, so it'll retain the width, but try to interpolate some of the surrounding white pixels to reduce the significance of the line, but ultimately, it stays the same size.  Upscaling this back to 100x100, you'll get a significantly wider line than you started with.  It'll be more "washed out", but it'll be certainly larger.

Most JPEG artefacts are small bits of noise, similar to this thin line.  Thus a downscale will effectively enlarge them.  It may make them less obvious, by blurring the surrounding colours in a bit (if it identifies the artefacts as non-edges), but if it doesn't actually eliminate them, they'll be enlarged in the upscale, and made more obvious via the sharpen filter.

By the way, I'm not sure what you're exactly referring to as haloing, so I'm unsure what you're exactly referring to, but the sharpen filter is a fairly basic one.

(This post was last modified: 29/11/2010 09:37 PM by ZiNgA BuRgA.)

(29/11/2010 09:37 PM)ZiNgA BuRgA Wrote:  Take an example: let's say wee have a 100x100 white image, with a single black vertical line, 1 pixel wide drawn down somewhere.
If wee downscale it to 25x25 using nearest neighbor, the line is either going to disappear (become 0 pixels wide) or stay the same width, depending on where the line is drawn.  If this position is random, according to probability, the line will most likely disappear in this case.
You've mentioned an average, which works in this case because nearest neighbor is non-discriminatory, as you've said.

Most JPEG artefacts are small bits of noise, similar to this thin line.  Thus a downscale will effectively enlarge them.  It may make them less obvious, by blurring the surrounding colours in a bit (if it identifies the artefacts as non-edges), but if it doesn't actually eliminate them, they'll be enlarged in the upscale, and made more obvious via the sharpen filter.

Yeah, but a black line is very different from the white background.  While the sort of artefacts you get from JPG generally differs far less from the background, so they will "meld in" better when blurred.

Wee're talking about tiny amount of JPG artefacts that are hardly even visible (which are likely to just get blurred to oblivion), not like massive amounts.  So yeah, pretty much, what I'm saying is, the JPG has nothing to do with the haloing you get at the end.

(29/11/2010 09:37 PM)ZiNgA BuRgA Wrote:  By the way, I'm not sure what you're exactly referring to as haloing, so I'm unsure what you're exactly referring to, but the sharpen filter is a fairly basic one.

Wikipedia ripoff - http://en.wikipedia.org/wiki/Ringing_artifacts

Quote:In signal processing, particularly digital image processing, ringing artifacts are artifacts that appear as spurious signals ("rings") near sharp transitions in a signal. Visually, they appear as "rings" near edges; audibly, they appear as "echos" near transients, particularly sounds from percussion instruments; most noticeable are the pre-echos. As with other artifacts, their minimization is a criterion in filter design.
Contents


Introduction
The main cause of ringing artifacts is overshoot and oscillations in the step response of a filter.

The main cause of ringing artifacts is due to a signal being bandlimited (specifically, not having high frequencies) or passed through a low-pass filter; this is the frequency domain description. In terms of the time domain, the cause of this type of ringing is the ripples in the sinc function,[1] which is the impulse response (time domain representation) of a perfect low-pass filter. Mathematically, this is called the Gibbs phenomenon.

One may distinguish overshoot (and undershoot), which occurs when transitions are accentuated – the output is higher than the input – from ringing, where after an overshoot, the signal overcorrects and is now below the target value; these phenomena often occur together, and are thus often conflated and jointly referred to as "ringing".

The term "ringing" is most often used for ripples in the time domain, though it is also sometimes used for frequency domain effects:[2] windowing a filter in the time domain by a rectangular function causes ripples in the frequency domain for the same reason as a brick-wall low pass filter (rectangular function in the frequency domain) causes ripples in the time domain, in each case the Fourier transform of the rectangular function being the sinc function.

There are related artifacts caused by other frequency domain effects, and similar artifacts due to unrelated causes.

Encoders refer to "haloing" as the more specific case of ringing as you see in the picture above, which look like bright halos around dark lines (and vice versa).

While "ringing" in general can refer to a lot of other things, like edge noise and JPG edge artefacts.



Your picture doesn't look bad in itself, but if you compare against the original, which has perfectly clean lines, you can see a big difference.

(This post was last modified: 29/11/2010 11:11 PM by Assassinator.)