rico wrote:
Moire (and color artifacts) can engulf the entire image in an obvious way, but an over-sampled image is also noticeable: I sense it as noise, or grain. If certain photogs and artists see over-sampling as Good Stuff, well, good luck. I prefer to start with a sampling of reality - hold the signal ambiguity - and add my own flavor of fantasy in post-processing. That's Art!

it's true adding noise can often make an image seem sharper but not using a low pass filter is not at all the same thing. the artifacts are in fact signal (a sampling of reality) just not the whole signal and you can't get the whole signal. while this means you aren't showing things accurately there are many cases where it comes closer to reality then simply throwing out all the high frequency information. luka's shots of the window blinds are a good example of this – low pass filtering makes it look like there are no blinds while showing the aliasing makes it look like their are blinds with wider shades than in reality. i personally like the unfiltered version in this case. unless you are batch processing it is pretty easy to see where the high frequency information will have a negative impact on downsizing and deal with it (or not) appropriately.

I took Luca's generated image and did couple of experiments on it.

* I replicated Luca's M9 script and run it on the image. It shows vertical lines in 40lp/mm, which in my opinion should not be there. The presence of these details indicates that the script is not resizing small details proportionally - it makes them larger. In real life picture, this script is a little bit too crispy for my taste. Luka's pictures look great with it, but not mine. For example, a round satellite dish showed jagged edge

* I tried imagemagick (used here by mpmendenhall before), because it has broad selection of resizing algorithms. I tried about six of them; I selected those that were recommended in brief internet research (blackman, bohman, hamming, mitchell...). They all converted the image pretty much with the same result with the 40lp/mm as a smooth gray area. In real life images these algorithms give similar results to each other and if followed by smart sharpen or USM, they give results that are somewhere between Luca's sharpening and ordinary bicubic followed by sharpening.

Lanczos:

Mirekt: I can asure you that latin(?) is not very commonly used in Sweden

Anybody got a GOOD solution for getting rid of CA in Photoshop? I usually just select the color using 'Color Range' and then modify the selection according to the situation. Is there an easier way?

denoir wrote:
Now look at the blinds. In the AA image we can barely see them (in fact had they been perfectly anti aliased we shouldn't see them at all) while in the plain image they are clearly visible - again, with the wrong frequency. But what is more important qualitatively - to show that there are blinds in the window or to show them at the right frequency? What possible improvement would a gray blur have been?


I can only admire your technical skills and efforts that you put into such experiments. Yes, the aliasing suggests the presence of blinds, but only because information is used that is not available from the image itself. In the majority of other images aliasing only garbles genuine detail, and your example does not make me appreciate aliasing any more than before. The improvement of a gray blur would have been: be silent instead of fabricating a story.

Toothwalker wrote:
Yes, the aliasing suggests the presence of blinds, but only because information is used that is not available from the image itself.

actually, no. the information is in the image, it just isn't complete. the blinds are actually a pretty weird example i'll admit though. regular patterns such as blinds usually present the worst possible outcome with regards to aliasing. more naturalistic scenes where information is not confined to a single frequency can often show accurate elements of the scene that would be lost if all high frequency info was thrown out.

Toothwalker wrote:
In the majority of other images aliasing only garbles genuine detail, and your example does not make me appreciate aliasing any more than before. The improvement of a gray blur would have been: be silent instead of fabricating a story.


that is quite an anthropomorphization of signal processing. the human visual system is actually quite good at taking partial information and creating an entire image from it and, if i recall correctly, does not use an anti aliasing filter itself in the time or spatial domain (there is a bit of "software based" filtering, but we still see some aliasing artifacts in the "real world"). for a landscape scene (no manmade structures with odd frequency distributions) it is entirely plausible that an image taken with aliasing would look more like the actual scene seen by being there than a filtered image would.

Thanks Paul. I guess I'm frustrated by the declining amount of actual engineering work I get to do on a regular basis so I'm taking it out on my photography

As for the rest, sebboh has just about covered everything that I'd say.

The point about how the human visual system works is very good. The actual information that comes from the eyes is minimal compared to the one that is approximated, extrapolated or outright fabricated by the brain to provide us with the visual experience that we have. It may not score high in fidelity from a signal processing point of view but at the same times it works quite well. It does an amazing job of providing us a comprehensive view that is built on very partial information and in the end it's much more useful than a system that pays attention to signal fidelity would be. It's not perfect and it can be fooled (optical illusions are a classic example), but on the whole it's a pretty good solution.

sebboh wrote:
Toothwalker wrote:
Do I really need to show you what happens if you downsize an image by just taking every Nth sample? I think you already know how aesthetically pleasing that will be.


but really it's not like anything below nyquist is unaffected by sampling frequency. there is a pretty severe phase amplitude interaction near (but below) nyquist. would you throw out that data too because it is corrupted? anybody who is serious about getting an artifact free signal will low pass filter well below the nyquist, choosing the appropriate sample rate to allow them to do this of course, in applications where this is possible. obviously photography is not such an application. everyone wants as much detail as is possible and in many cases adding false high frequency details can, i suspect, be shown to be preferred by a significant majority of photo viewers for many types of images.

I think there is nobody here that denies that there are trade-offs. Tolerating a little more aliasing so that more detail is passed elsewhere ... fine. There are dozens of filters to choose from. Denoirs sine wave example of picking every Nth sample followed by the remark that he has not heard any convincing arguments against it is however far beyond any reason.

Anybody got a GOOD solution for getting rid of CA in Photoshop? I usually just select the color using 'Color Range' and then modify the selection according to the situation. Is there an easier way?

For lateral CA, applying lens profiles in ACR/LR works well for me.

Toothwalker wrote:
I think there is nobody here that denies that there are trade-offs. Tolerating a little more aliasing so that more detail is passed elsewhere ... fine. There are dozens of filters to choose from. Denoirs sine wave example of picking every Nth sample followed by the remark that he has not heard any convincing arguments against it is however far beyond any reason.


Hint: Try actually reading what I wrote instead of fabricating statements never made.


denoir wrote:
Now in a real world image things are more complex where you have a blend of multiple spatial frequencies. And in effect you'd be blending the poorly reconstructed signal with a properly reconstructed one. So which is worse, sabotaging the good one or throwing away information from the poor one? I don't know and I have so far not heard any convincing arguments either way.


Actually I don't stand by that statement. My AA-through-diffraction experiment convinced me that throwing away information from the poor one is worse.

sebboh wrote:
that is quite an anthropomorphization of signal processing. the human visual system is actually quite good at taking partial information and creating an entire image from it and, if i recall correctly, does not use an anti aliasing filter itself in the time or spatial domain (there is a bit of "software based" filtering, but we still see some aliasing artifacts in the "real world"). for a landscape scene (no manmade structures with odd frequency distributions) it is entirely plausible that an image taken with aliasing would look more like the actual scene seen by being there than a filtered image would.

Indeed the human visual system is quite remarkable. It may also work differently on a real subject than on an image of the same subject. A well-known example is a person who looks up against a building and thinks nothing in particular, but strongly objects to a picture with the same perspective because he notices converging verticals.

Concerning the landscape, we can just look at an example.

With aliasing:

Toothwalker wrote:
Concerning the landscape, we can just look at an example.

With aliasing:

denoir wrote:
Now, shall we stop being silly or is this discussion over?


I like that suggestion. You made your choice and I made mine.

Toothwalker wrote:
Indeed the human visual system is quite remarkable. It may also work differently on a real subject than on an image of the same subject. A well-known example is a person who looks up against a building and thinks nothing in particular, but strongly objects to a picture with the same perspective because he notices converging verticals.

indeed there are many differences, but there are a number of pattern completion algorithms implemented by the brain the work the same looking at a picture as they do looking at the world around you.

Toothwalker wrote:
Concerning the landscape, we can just look at an example.

With aliasing:

sebboh wrote:
or if we could look at something more useful like a series of different examples with different levels of filtering in different configurations rather than a caricature of aliasing. i could have done something equally as silly and shown an example where the shot without aliasing was just grey.

edit: luka beat me to it.

Yes, one can show an example of a shot without aliasing that is just gray. But at this point I am just reacting to statements that aliasing as such might be beneficial. The fact that throwing away signal below Nyquist makes an image soft is trivial. At this point we have a landscape where aliasing clearly ruins it, and another landscape with aliasing that looks much better - but where we do not know whether the aliasing actually contributes to the appreciation, or makes it look more like the actual scene as you wrote. (I am quite sure it does not, but realize that I cannot prove this.)

Toothwalker wrote:
Yes, one can show an example of a shot without aliasing that is just gray. But at this point I am just reacting to statements that aliasing as such might be beneficial. The fact that throwing away signal below Nyquist makes an image soft is trivial. At this point we have a landscape where aliasing clearly ruins it, and another landscape with aliasing that looks much better - but where we do not know whether the aliasing actually contributes to the appreciation, or makes it look more like the actual scene as you wrote. (I am quite sure it does not, but realize that I cannot prove this.)


your last statement in parentheses is what bugs me about theoreticians and people on the internet: no desire to find out if theory works in practice, just faith. simple mathematical truths do not always translate well into complex systems where all the functions and variables are unknown. if you want to know the truth you need to actually do the experiment. if you just want to be right, well you are. any entry level textbook on signal processing will tell you so.

the answer is obvious, controlled double blind experimentation to determine whether aliasing can be beneficial or is always detrimental. since there are certainly a large number of techniques for filtering, it would probably be beneficial to look at our (still rather limited knowledge of) what filters the brain applies at different points in the processing stream to cherry pick what might work well with them...

Luka,

Been away from the discussion for a bit. Personally I'm leaving most of it at the point of aesthetics. Stepping back into the camera AA filter analogy, while I'd seriously consider removing the AA filter on my camera bodies if it were practical (out of curiosity if nothing else) I do know from looking at lots of landscape shots from other people's AA-less cameras that the "false detail" some people fawn over (particularly in grass and fur) actually look like a "crunchy" artifact mess to me. So clearly there are different tastes, and I'd expect them to depend on the subject as well.

What I really wanted to post about though was that the test setup for your aperture experiment was awesome!

Ken

You are sometimes driving pretty far from reality ppl.

Who seriously cares why M9 doesnt have AA, when you can produce such amazing photos as denoir? (well, he shows its possible, its not like everyone can.. usually they cant, but thats not the point )

I dont care if its cause Leica doesnt have money for AA (ehm.. sure, I believe that given price of camera ), or simply didnt want to put it there. What matters is, that results are exceptional, and from my point of view, best from all current full-frames. And that moiré is actually quite easy to fix and if you didnt know it was there, you wont find it.


So maybe, more constructive posts similar to "how to resize properly", "how to get rid of moiré". Then fighting about why M9 doesnt have AA.

I have one question, whats best way to upsize photo?

I tried Geniuine Fractals, definetly better than just plain PS, but nothing exceptional.

And as someone asked, what about CA, any super tricks?

Mescalamba wrote:

I have one question, whats best way to upsize photo?

I tried Geniuine Fractals, definetly better than just plain PS, but nothing exceptional.

I've made quite large prints from standard consumer camera files (exhibition backgrounds and banners, not fine art prints), largest upsizing has probably been 6x4 metres print from 16 mp original file.

Earlier I tried different fractal softwares but imo they weren't worth the effort.

My current method is opposite the step-by-step sharpening used for websize fotos. When the goal for websize photos is to maintain some of the detail from original pixels, the goal in upsizing is to spread the original pixels to wider area to make the illusion of image with more detail then there originally was. And to try keep the lego-effect as small as possible. In photoshop I get the best results by upsizing and downsizing the photo so that the pixels never get multiplied by easy numbers. So forexample I might increase the imagesize by 70% and then decreasing by 30% and repeating until the required size is achieved. The numbers really depends on the resizing factor and the imagetype. I usually don't sharpen image at all when upsizing. Sometimes the original might need some pump but usually I just leave it that.

denoir wrote:
The point about how the human visual system works is very good. The actual information that comes from the eyes is minimal compared to the one that is approximated, extrapolated or outright fabricated by the brain to provide us with the visual experience that we have. It may not score high in fidelity from a signal processing point of view but at the same times it works quite well. It does an amazing job of providing us a comprehensive view that is built on very partial information and in the end it's much more useful than a system that pays attention to signal fidelity would be. It's not perfect and it can be fooled (optical illusions are a classic example), but on the whole it's a pretty good solution.

Does anyone know more about techniques such as this (also developed by NASA I believe): http://dragon.larc.nasa.gov/retinex/pao/news/
Looks interesting. I only see haloing in one or two of the examples.
I believe the technology is based on what is discussed in the video linked to by douglasf13: http://www.youtube.com/watch?v=ALfiTDYLtAQ (starting from 47 minutes the relevant parts start)


@mescalamba: I believe PTLens is one of the best pieces of software to correct CA. If I remember correctly, it's one of the few tools that can correct for non-linear lateral CA. When I correct lateral CA in an image, often the CA that disappears in one place pops up in another place where there was no CA before. I don't know if this is due to different focus depths or non-linear CA, but the tools in applications such as Lightroom and Photoshop seem to be relatively blunt instruments.

Sami Ruusunen wrote:
I've made quite large prints from standard consumer camera files (exhibition backgrounds and banners, not fine art prints), largest upsizing has probably been 6x4 metres print from 16 mp original file.

Earlier I tried different fractal softwares but imo they weren't worth the effort.

My current method is opposite the step-by-step sharpening used for websize fotos. When the goal for websize photos is to maintain some of the detail from original pixels, the goal in upsizing is to spread the original pixels to wider area to make the illusion of image with more detail then there originally was. And to try keep the lego-effect as small as possible. In photoshop I get the best results by upsizing and downsizing the photo so that the pixels never get multiplied by easy numbers. So forexample I might increase the imagesize by 70% and then decreasing by 30% and repeating until the required size is achieved. The numbers really depends on the resizing factor and the imagetype. I usually don't sharpen image at all when upsizing. Sometimes the original might need some pump but usually I just leave it that.


i'm actually curious about what people have found works best to sharpen for print and how that changes depending on whether you are upsizing, downsizing, or printing at original size? it's a lot more expensive to try out different techniques for print than it is for lcd viewing.