AhamB

Interesting links and thanks for PTlens tip, I will try it.

That Retina thing, dunno.. I can do same from one RAW and mix of PS/Photomatix work. Maybe some noise smoothing involved, if shadows are pushed too far. I know that, cause I was actually doing that a lot (m4/3s needs a bit of help to get DR right). Most impressive technique in this regard I saw was Zero Noise, but that SW kinda vanished..

Mescalamba wrote:
Most impressive technique in this regard I saw was Zero Noise, but that SW kinda vanished..

It isn't free, but you might look at PhotoAcute to do some of the things Zero Noise would do. It is a RAW multiple exposures in to single RAW file out software. The documentation is a bit weak, but there is a free trial that watermarks so you can experiment with it extensively. I've tried it with simply averaging multiple base ISO shots to get effectively a lower ISO and lower noise RAW file on m43. It can do many fancier things as well. What is nice is that it is RAW to RAW.

Ken

sebboh wrote:
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.

I hardly ever print, but what I've picked up over the years is that print requires much more agressive sharpening -- think obvious sharpening halos when inspected on screen. There are some articles on LL about it but I'm sure there must be other resources. Some sharpening programs like Nik Sharpener have an option to choose the output medium, print size and viewing distance and the level of sharpening is adjusted accordingly.

@Ken: Does the PhotoAcute software output a DNG or something? I think the Zero Noise tool worked with TIFF files.

Yep, PhotoAcute works with DNG files, both in and out. Some functions (like noise averaging) do essentially nothing to the RAW pixels spatially while others are clearly doing various forms of interpolation and require lens and camera profiles to work.

Ken

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.

I'd recommend some reading from Bruce Fraser and Jeff Schewe.

http://www.amazon.com/World-Sharpening-Photoshop-Camera-Lightroom/dp/0321637550/ref=pd_sim_b_2
http://www.luminous-landscape.com/videos/camera-print.shtml


Sharpening in PhotoKit Sharpener (with multiple options for various types of print devices from inkjet to offset) and in LR (optimized for inkjet, I believe) were inspired by the same people.

sebboh wrote:
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.

my basic rule of sharpening for prints are:

large fine art print enlargements (coated): no sharpening at all
large prints with uncoated paper or some other material where the ink spreads: mild sharpening

medium sized (magazines and medium paper photos) coated and uncoated: mild sharpening

small fine art prints or magazine paper (coated): mild sharpening
small prints for uncoated material or newspaper where ink spreads: heavy sharpening

ofcourse the sharpening differs because of the style of the photo and where it is intended to use, ie prints which are viewed close can not be sharpened so much than those viewed from the distance

Bump

thanks ahamb, mirek, and sami. i'm mostly interested in wall mounted prints in the 8 x 10 and up range. i'll have do more reading and more experimenting.

carstenw wrote:
Bump

Since this thread has been kindly bumped, I'll just add a small demo of color moire. Test apparatus includes the Canon 1Ds (weak AA filter), CZ S100 C/Y (quite sharp @ f/5.6), and flash. The test pattern is Postscript source onto laser printer with one square black per 2x2 cell, aligning to the 1Ds Bayer layout with 99% accuracy at test distance. The pixel drift is discernible in the changing color moire exhibited below. First image shows the overall scene at test distance with critical focus:



Conclusion: passing high frequencies to the sampling stage will induce false data that, in the case of a moire pattern, may cause an error both global and uncorrectable.

Thanks Rico. I had objected to the statements in this thread about color artifacts only being on the pixel level as I knew that to be false but didn't have the time or energy to do a demonstration. Great illustration!

Ken

rico wrote:
Conclusion: passing high frequencies to the sampling stage will induce false data that, in the case of a moire pattern, may cause an error both global and uncorrectable.

That is not disputed, but your test shots are a really good argument for not having an AA filter.

I hope you don't mind if I use crops from your test shots. As with the blinds example it comes down to a basic choice (at least in theory). The first crop shows the actual pattern as it was shot at a distance that guaranteed the pattern to be within the Nyquist limit. Shot A shows a "false" pattern (much lower frequency than the actual pattern) and shot B is what a theoretically perfect AA filter would do. Since the spatial frequency is too high to be represented accurately a perfect AA filter will just result in a gray average:

denoir wrote:
That is not disputed, but your test shots are a really good argument for not having an AA filter.


Umm... A wild rainbow of color from a gray scale target is a good argument for no AA filter?

I agree, there are good and bad reasons for an AA filter, but you went off the deep end with your opinion on this one. Converting grey to a kaleidoscope is clearly not a good thing. Furthermore, your point that moire color artifacts are only visible at the pixel scale has been demonstrated to be completely wrong. Of course this isn't a "real world" example, but you are completely twisting basic observation and common sense.

You'll have to give it a rest on this one I'm afraid! Leave it at "matter of choice and preference" and preserve your credibility.

Ken

kwalsh wrote:
Umm... A wild rainbow of color from a gray scale target is a good argument for no AA filter?

Preserving the information about the dotted pattern is a good argument for no AA filter. The color artifacts you can get rid of in PP in most situations but there is no way of reconstructing information you have thrown away.

And actually, now that you mention it, the wild rainbow of colors is indeed a good argument for no AA filter. That was shot with a camera that has an AA filter. I mean, why bother? You've killed off detail with it and you still get the moire.

A 5DII with a strong AA filter would have produced a rainbow pattern there as well with little improvement over what would be accomplished with a simple PP operation after the fact but at a high cost in loss of detail. Yes, it would have been weaker but in a case like this the strength of the rainbow pattern is inconsequential as long as it's there and visible.

Furthermore, your point that moire color artifacts are only visible at the pixel scale has been demonstrated to be completely wrong.

Err.. I have no idea what you are talking about. I don't recall saying any such thing nor have I been able to find any such statement by anyone in the past pages and definitely not any made by me. If you have a rainbow pattern that appears across a large region, of course you will see it in a resized image unless you fix it in Lab mode or excessively low pass filter the image. What of it?

You'll have to give it a rest on this one I'm afraid! Leave it at "matter of choice and preference" and preserve your credibility.

Ken, I'm disappointed in you. I thought we had agreed to skip this type of cheap personal attacks.

I guess he's in love with AA filters. Love is blind they say.

I think it's an interesting thread and deserves to be approached maturely and as scientifically as can be.

More samples!

denoir wrote:
kwalsh wrote:
Umm... A wild rainbow of color from a gray scale target is a good argument for no AA filter?

Preserving the information about the dotted pattern is a good argument for no AA filter. The color artifacts you can get rid of in PP in most situations but there is no way of reconstructing information you have thrown away.


You can recreate any sort of false data you want or remove any false data you want anywhere in the image. Whether you want to be desaturating portions of the image to deal with color moire or start adding detail or sharpening to regions is up to the photographer. In some cases one is harder than the other or more frequently necessary than the other. I think this was already covered pretty thoroughly.

That said, I have already agreed with you that in many cases the false information looks nice. In others, to my taste (and others) it looks like crunchy garbage. That was my main point here. Leave it at that - sometimes good for some, sometimes bad for others.


And actually, now that you mention it, the wild rainbow of colors is indeed a good argument for no AA filter. That was shot with a camera that has an AA filter. I mean, why bother? You've killed off detail with it and you still get the moire.


Silly point. You get global color issues in more situations without the AA than you do with. It is a trade off as to what you shoot and what post-processing you want to deal with. Again, sometimes useful sometimes not.


Furthermore, your point that moire color artifacts are only visible at the pixel scale has been demonstrated to be completely wrong.

Err.. I have no idea what you are talking about. I don't recall saying any such thing nor have I been able to find any such statement by anyone in the past pages and definitely not any made by me.


This post:

http://www.fredmiranda.com/forum/topic/1038448/14#9874135

Specifically this quote:


Color moire occurs when you have a lens that resolves detail higher than the nyquist frequency at very high MTF and it's a high contrast, high frequency subject and you get an unfortunate alignment with the CFA. An AA filter smears the light across the CFA quadruplets (GRGB). Without an AA filter and with the conditions I mentioned above you can get for instance that GRG records black (zero signal) while B gets a full blast of white. A perfectly AA filtered sensor would in the case of such an edge record G=0.5, R=0.5, G=0.5, B=0.5. In the case without the filter you get G=0, R=0, G=0, B=1. Then during raw development the demosaicing algorithm calculates a linear combination of the colors for each of the four pixels and ends up with something blue.

That's the basic mechanism. The point however is that this is at the local pixel level and the color artifacts you see are a few grouped pixels with wildly varying colors. You can't get a color shift from it. It just looks like very high frequency colored noise.


You appear to state color effects are only on a local pixel level and can not be seen as color shifts on a more global scale. That's not true. Apologies if I misread or misunderstood what you wrote there.


You'll have to give it a rest on this one I'm afraid! Leave it at "matter of choice and preference" and preserve your credibility.

Ken, I'm disappointed in you. I thought we had agreed to skip this type of cheap personal attacks.


It wasn't meant as a personal attack! Apologies if it came off that way, "credibility" was probably not the appropriate word.

I just meant it reads a bit silly when you completely ignore the OP's point about obvious and gross color artifacts to make the same point about pleasing false detail you've already expressed repeatably in the thread. I think we know where you are coming from on that, and I agree with you that in some cases it is useful. Really though, reviving a dead thread and avoiding commenting on the most recent information in it only to rehash a point already beat to death didn't seem to be adding much to the discussion.

That said I probably added even less and should have just left it to scroll off the front page again...

Ken

kwalsh wrote:
You appear to state color effects are only on a local pixel level and can not be seen as color shifts on a more global scale. That's not true. Apologies if I misread or misunderstood what you wrote there.

I guess I should have been clearer in that post but it was in context to a specific real world sample where there was no uniform pattern. I don't dispute that you can create color shifts if you have an unfortunate alignment of an artificial high contrast pattern and the Bayer grid. There's a bunch of stuff you could do if you are careful to select and align a pattern - you could for instance make a dotted pattern to appear as a solid R, G or B color. You are however extremely unlikely to encounter such an arrangement in a real world shot.

In most real world situations color artifacts are a local problem similar in nature to the color noise at high ISO and dealt with the same way. There are of course exceptions in case of artificial patterns that occur in real world images (fabrics for instance are a typical example) where the problem becomes global.


It wasn't meant as a personal attack! Apologies if it came off that way, "credibility" was probably not the appropriate word.

Yeah, well, I might add that responding to something by saying "silly point" isn't very conducive to maintaining a civil discussion. I'm afraid however that I won't take the bate and match the rudeness as I have no interest in wasting time with a pointless escalation of throwing insults. But as I said, I'm disappointed as I thought that we had reached an understanding there.

I just meant it reads a bit silly when you completely ignore the OP's point about obvious and gross color artifacts to make the same point about pleasing false detail you've already expressed repeatably in the thread.

We've discussed color artifacts repeatably as well. The interesting part of the OP's test was to me the example of the false detail which was more direct and more elegant than my blinds example.

As mentioned already by Ken, you made the falsifiable statement "this is at the local pixel level and the color artifacts you see are a few grouped pixels with wildly varying colors. You can't get a color shift from it. It just looks like very high frequency colored noise." Such statements are, of course, in the proper spirit of science, and I voiced my disagreement shortly afterward in that other thread:

http://www.fredmiranda.com/forum/topic/1038448/15#9874592

The results from my test are pretty conclusive, but some follow-up is in order.

Firstly, the Canon 1Ds has an AA filter and, to the extent that it permits aliasing, is deficient in its role. A more effective AA filter would have made this experiment with the 1Ds impossible.

Secondly, your point about the aliased pixels in one of my resized test images is quite valid. The resizing algorithm of the Gimp is apparently susceptible to that 2x2 pattern at certain reductions. The scale factor was 0.25x and caused a pixel pattern that is both pretty and perfectly false. It is a testiment to the insidiously nature of false data that I missed it completely (although my emphasis in the report was aliasing of the color variety). Note that a slightly different factor of 0.20x causes false data both different and considerably more disturbing:



I have no idea what resizing algorithm is used, but I recall no moire errors coming from this direction in past use. The scale factor of 0.25x for this pattern is clearly an issue, but the resizing algorithm has an effect, too. Note the various renditions with the selectable "filters" of ImageMagick:



Finally, the problem of color and pattern alias error affects my photography because I shoot a lot of fabric in full color with studio lights. A sample:

I'll throw in thanks for sharing the PP ideas. Robert

BTW, those "resizing algorithms" are called Reconstruction Filters.

Hehe, this was my job as a software developer for awhile at NewTek where we made and improved an application called Lightwave 3D. It might help to know the term in case someone wants to look something up in their research.

Could anybody care to elaborate on addressing CA issue using gaussian blur in Alpha channel? Obviously, creating Alpha Channel and blurring it all will hardly do any good..