Bad times for LEICA S- guess they will be severely hurt

[ErikKaffehr]

Hi Eric,

I don't think I have seen your point about smaller pixels. I have stated for many years that smaller pixels are good, not least to reduce aliasing. The reason this is dicussed here is that the Leica has relative large pixels  and very good lenses. So it is really a recipe for aliasing.

MFD photographers have opted for small pixels for a long time, and one reason was to get rid of colour moiré, one form of aliasing.

If you shoot an 80MP back I guess that you get rid of colour moiré some around f/11 (didn't do the math, just an estimate). But, any well corrected MF lens will be sharpest between f/5.6 and f/8. With f/11 you loose a bit.

Sharpening comes into play, as long as you haven't stopped down to far small contrast can be enhanced by sharpening.

Bart van der Wolf has a lot of good stuff on this.


Best regards
Erik

I'm not sure one can get away with such a presumption that with more pixels the final image size would stay the same with more and smaller pixels.  Rather, with more pixels people will crop more, print bigger, or just zoom into 100% and pixel peep.  I don't think Canon can sell the new 5Ds with more and smaller pixels and tell their customers that they can only expect better quality if they keep the prints the same size as with their 5Dmk2.

You yourself claimed that you'd be limited by diffraction in the posted image if you stopped down to f/16 to avoid aliasing.   I doubt with a p45 its enough to spoil an image more so than aliasing since I shoot my 80mp camera frequently at f/16 and do fine, but in any case you've just proven my point about smaller pixels.  If you were not willing to stop down to f/16 due to diffraction concerns in the above image and a p45, then you would really wouldn't be happy with a similar camera with your own optimum pixel pitch where you would be limited to a much larger aperture, than you are now.   

[EricWHiss]

Yes, Erik, you missed my point.
My ideal pixel is a lot bigger than your 3um value at least 2x bigger.  And I wouldn't be surprised if my ideal sensor size would be different from yours or a lot of people for that matter. I'd like to see something bigger than what we have currently. 

I haven't seen a whole lot of them but haven't seen any color aliasing in any of the Leica S files I did look at.  Have you?  But surely there is some but probably the users are working around it just like the p25 owners did. I have never felt color aliasing was as important a concern as you have all along. 

On the 80mp backs, you can see diffraction start at f/8 but its not really enough loss to worry about until past f/16.  I have Imatest data for all of my lenses from wide open to f/18 or so - you can see it but it's very small until just before f/16. It's not that important really - the most affect area is the peak of the focus which also is much sharper than the areas still in DOF but not at the focus peak.  I doubt that anyone would notice the difference between f/11 and f/13 and few would notice between f/13 and f/16 and besides you can recover some diffraction loss with software so I don't worry about it.   I shoot often at f/13 and f/16 with the camera - works brilliantly.   

ps you might enjoy imatest software if you haven't already a copy.

[ErikKaffehr]

Dear Sir,

I have had Imatest since it's inception. I have also published a lot of data using Imatest, which you have bin happy to criticise. You memory seems to be short.

Regarding pixel size, here is an analysis Jim Kasson published. It is based on the Otus, a quite demanding lens and I think he used it at f/5.6.
- He found that at 1.25 microns the OLP filter was not needed
- There would be no or little visual difference between 1.25 micron and 2.5 microns
- Five microns (with OLP filtering) was still good

You find it here: http://blog.kasson.com/?m=201406

The image below shows one of Norman Koren's test charts shot at 1/50 magnification. Top is 3.8 micron Sony Alpha 77 (probably OLP filtered), and below is the P45+ used with a Planar 80/2.8 at f/5.6. The Sony Alpha still has some aliasing but the P45+ has a lot. The Sony was shot at f/8 in this case, so it has some more diffusion from diffraction. The reason I used f/8 on the Sony is that the lens used was a 16-80/3.5-4.5 zoom and I wanted to have it a bit stopped down, but regarded the Planar be near optimum in resolution at f/5.6. What you can keep in mind is that any information on the right of the red line is fake information. Ideally that area should be smooth gray! Ideally, the lens + OLP filter + sensor combo would have MTF of zero at Nyquist. In that case there would be no aliases and the image could be sharpened to increase MTF at high frequencies.




That Planar can resolve something like 150-180 lp/mm on ADOX CMS, albeit at very low contrast.

MTF-curves maesured with Imatest are shown here (both at f/8):


Would you be interested in the issues involved the article may be worth reading: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/78-aliasing-and-supersampling-why-small-pixels-are-good

Personally, I feel that when there is a conflict between proven theory and experience there is something wrong with either the experience or the interpretation of the theory.

Best regards
Erik

Yes, Erik, you missed my point.
My ideal pixel is a lot bigger than your 3um value at least 2x bigger.  And I wouldn't be surprised if my ideal sensor size would be different from yours or a lot of people for that matter. I'd like to see something bigger than what we have currently.  

I haven't seen a whole lot of them but haven't seen any color aliasing in any of the Leica S files I did look at.  Have you?  But surely there is some but probably the users are working around it just like the p25 owners did. I have never felt color aliasing was as important a concern as you have all along.  

On the 80mp backs, you can see diffraction start at f/8 but its not really enough loss to worry about until past f/16.  I have Imatest data for all of my lenses from wide open to f/18 or so - you can see it but it's very small until just before f/16. It's not that important really - the most affect area is the peak of the focus which also is much sharper than the areas still in DOF but not at the focus peak.  I doubt that anyone would notice the difference between f/11 and f/13 and few would notice between f/13 and f/16 and besides you can recover some diffraction loss with software so I don't worry about it.   I shoot often at f/13 and f/16 with the camera - works brilliantly.  

ps you might enjoy imatest software if you haven't already a copy.

[torger]

I haven't seen a whole lot of them but haven't seen any color aliasing in any of the Leica S files I did look at.  Have you?  But surely there is some but probably the users are working around it just like the p25 owners did. I have never felt color aliasing was as important a concern as you have all along.  

This "being without AA filter is a feature" is about photographer psychology rather than image quality. Of course there's aliasing. But there's also pixel peep joy. Photographers rarely admit it but we just love to pixel peep, and there's a special type of satisfaction of 100% crops that are so sharp that they cut you.

If we would care about real image quality we would not worry about the 1-2% resolution loss of an AA filter, but rather about spurious and unpredictable aliasing.

I've had a long long discussion about 5DS vs 5DS R recently, and even when I show examples of moire and aliasing from the few 5DS R shots out there people still want the R for the more pleasing pixel peep, it's just more fun. I don't blame them, I have a hard time resisting myself, but one should be aware of that it's an illusion that it's about image quality.

I don't think people generally "work around aliasing" (shaking the camera, shooting at f/22, shooting out of focus), they just ignore it. In most cases it's really subtle and clients won't notice. Only specialists on digital image quality will notice, unless you have some really bad moire (which indeed is quite rare).

When we shoot medium format we of course have no choice as there are only AA filter-free sensors out there. Then you learn to not care, or even regard it as a feature. I myself shoot mostly at f/16 with my 6um sensor, I've found that after proper sharpening workflow it yields better image quality with no significant detail loss compared to f/11, thanks to reduced aliasing. But yes, pixel peeping is not as fun when you've softened the image with some diffraction... but as everyone knows pixel peep sharpness only matters for pixel peep entertainment, as soon as you publish the image in any form print, web etc, the image is scaled either up or down and it just doesn't matter, you're actually better off with a softer more scaling-friendly image.

Of course aliasing is just one aspect of image quality, like noise, overall color etc. It's a matter of taste how important you think it is to register the correct color of small details. As you often scale down images for publishing it's often not that important, unless it forms moire. So I agree it's not too hard to live with, but I do understand those that consider it to be an issue, just like some think having "only" 12 stops of dynamic range is an issue.

[paratom]

The other side is there might be many - including me- who have used sensors without AA-filters for many years and have not been disturbed by aliasing. I am sure if I search for it I will find it in some images.
On the other side I am not yet convinced if you see the difference between AA-filtered and non-AA-filtered only when pixel peeping.
And how much can we close the "gap" with sharpening? I allways feel sharpening is adding artificial information to an image which has not been there in reality. Isnt it a bit like a simulation of missing information?
The approach of the Pentax K3 seems great, to have a function to slightly shake the sensor and to be able to choose.
I am not arguing that AA-filter might make sense.

[Telecaster]

And how much can we close the "gap" with sharpening? I allways feel sharpening is adding artificial information to an image which has not been there in reality. Isnt it a bit like a simulation of missing information?

"Sharpening" is really just selective contrast enhancement. If poorly/over done it can add objectionable artificial detail, like haloing.

Aliasing isn't just about moiré effects…it does add artificial spatial detail too. This contributes to the crunchy sharp full res look that, as torger notes, so many of us get off on. A sensor with a large enough photosite count to mitigate aliasing is likely to be a letdown in this respect 'cuz that fake detail will go away.  

Edit: Edmund (just below), good points. I'm not arguing that non-AA cameras/backs offer no additional real spatial detail. Just that existing ones, at least those I'm familiar with, are offering other stuff too.

-Dave-

[eronald]

Aliasing isn't just about moiré effects…it does add artificial spatial detail too. This contributes to the crunchy sharp full res look that, as torger notes, so many of us get off on. A sensor with a large enough photosite count to mitigate aliasing is likely to be a letdown in this respect 'cuz that fake detail will go away.  

-Dave-

It's not so simple. RB, and G have substantially different spatial frequencies in a Bayer sensor, and also different spatial frequencies if you measure horizontally or diagonally to image orientation. Look at the distance between two nearest G's and two R's in the Bayer array fragment below.

RGBGR
GBRGB
RGBGR

So either you really kill all the resolution of the G to avoid aliasing in the R and B or you will get the chromatic sparkling of false color. And this will be especially the case in eg. hair where thin lines change direction.

Now if you kill all these nasty frequencies by attenuating way below the diagonal  R and B frequency you have effectively divided your spatial resolution by .... look at the diagram above.

And there is the also the interesting detail that usually filters don't chop, they attenuate, so you have some progressive attenuation which blurs your image even more in the lower frequencies. Oh, and also I would expect some contrast loss and diffusion in *any* filter. All of the above possibly explain why digital backs that have no real AA filter have such a sharper look than the filtered SLRs: They really are sharper.

We can debate this, but before people explain to me as usual that I am not an "imaging processing expert", I would prefer that they show either simulations that take Bayer into account, and a realistic filter model (pun intended) or just write down the formulas taking the various sensel distances into account.

Edmund.

[torger]

With the 5DS and 5DS R the AA filter vs no AA filter question is as hot again as when the D800/D800E was released. I think it's interesting. I use AA-filtered cameras in the 135 range, and AA-filter-less cameras in MFD obviously. It should be said that a microlens-free sensor like my 50 megapixel Kodak with very sharp tech lenses will alias a bit more than a modern microlens-equipped sensor with fuzzier lenses.

Anyway, my taste has turned more towards the AA-filtered look the more I've worked with raw conversion on the software level. I've got to see how much that we actually make up during demosaicing. We make up a lot with AA-filtered sensors too otherwise we would just lose too much resolution, but there's considerably less problems. Afaik AA-filters today are not just random blurrers but carefully crafted filters designed to blur as little as possible while removing as much aliasing as possible, I guess that design is what makes them reversible with an extra layer like done with the D800E and now 5DS R.

Just knowing that details are not as correctly rendered as they could be just for 1-2% resolution gain does not feel right. On the other hand, when I pixel peep my tech cam lenses when shot at f/11 with my 6um pixels there I do get that sharpness rush I think everyone get. I doesn't help that I know that that sharpness does me no good when I print or scale, it's still a special kind of satisfaction. So I won't blame anyone that trades that satisfaction for some aliasing artifacts, but I think it's quite easily demonstrated that it's not about real image quality in the final output, but rather the opposite.

My taste towards AA-filtered look has grown even stronger lately when I've started to make prints with an own printer, as I've seen even less reasons to have that pixel peep sharpness when I scale for printing, rather the opposite, in the rare event of having to scale up an aliased image will just make it look "digital". Having a sharper pixel peep could have a value if you would publish the image on a medium where the square pixels in the image map exactly to square pixels in the output medium, but that is never the case. There's always scaling to match, and then aliasing does you no good.

It is an interesting question if removing the AA filter has a global effect, ie if you scale down the image so there's no resolution advantage, would you still see a difference (assuming we don't have visible moire in the subject). I've got the sense that there is a belief that there is indeed a "global advantage" in image quality of not having an AA filter, but I'm almost 100% positive that it's a myth (the only reason I would see is that AA filter would somehow disturb the efficiency of the color filters, but I haven't heard or read that there would be such a problem). I haven't done A/B testing though, there are not many cameras out there that allow that.

About that sharpening creates artificial detail, it depends on how you see it. The bayer+demosaicing process is softening in itself, and diffraction too. Counteracting those effects I would not say creates artificial details, ie you don't create any details that are not there you just enhance contrast. Aliasing creates false details for real though, in the form of small false colors and patterns around small details, which result in moire if the small details have a repeating pattern.

In landscape images moire occurs rarely, but false colors around small details occurs often. However with a 50 megapixel camera or so those details are so small so they generally become invisible in most publishing forms, and even if they are visible they're so tiny that the eye don't register that the color is wrong. So in practice it's certainly not a big issue for a landscape photographer like me. But I also know that the sligthly less sharp AA-filtered image would also not result in meaningful loss in sharpness, and then it for me feels better to rather trade (reasonably) accurate details for a visually reversible softening.

To me being without AA-filter on my MF camera is not a feature, but rather "a minor bug" I as a landscape photographer can live with.

[ErikKaffehr]

Hi,

I hope this example explains some issues. Just to say, this is an intended test.I found a subject that shows interesting issues and went back to shoot it with different cameras. In this case tripod position and focal length was fixed but different sensors were used. So this demonstrates the effect of pixel pitch and OLP filtering.

The image below was shot with a macro Planar 100/3.5 on a P45+. In this case the lens outresolves the sensor and has no OLP filter. Pixelation and moiré occours:


The next image was shot with Sony SAL 70-400/4-5.6G lens at 100 setting and f/8 on a Sony Alpha 77 with 3.8 micron pixels and probably some OLP-filtering:


Finally, the same image downsized to same image size as delivered by the P45+:


What I see here is that the P45+ image has both moiré in the blinders on the right side and some colour artifacts on the "Centralstationen" sign, it could be argued that there is some pixelation.

The center image shows the Alpha 77 image at actual pixels, clearly, the image is much cleaner but also quite soft.

The third image show the Alpha 77 image downscaled to 6.8 micron resolution. It may need some additional sharpening, but it shows more true detail than the first image.

Best regards
Erik

With the 5DS and 5DS R the AA filter vs no AA filter question is as hot again as when the D800/D800E was released. I think it's interesting. I use AA-filtered cameras in the 135 range, and AA-filter-less cameras in MFD obviously. It should be said that a microlens-free sensor like my 50 megapixel Kodak with very sharp tech lenses will alias a bit more than a modern microlens-equipped sensor with fuzzier lenses.

Anyway, my taste has turned more towards the AA-filtered look the more I've worked with raw conversion on the software level. I've got to see how much that we actually make up during demosaicing. We make up a lot with AA-filtered sensors too otherwise we would just lose too much resolution, but there's considerably less problems. Afaik AA-filters today are not just random blurrers but carefully crafted filters designed to blur as little as possible while removing as much aliasing as possible, I guess that design is what makes them reversible with an extra layer like done with the D800E and now 5DS R.

Just knowing that details are not as correctly rendered as they could be just for 1-2% resolution gain does not feel right. On the other hand, when I pixel peep my tech cam lenses when shot at f/11 with my 6um pixels there I do get that sharpness rush I think everyone get. I doesn't help that I know that that sharpness does me no good when I print or scale, it's still a special kind of satisfaction. So I won't blame anyone that trades that satisfaction for some aliasing artifacts, but I think it's quite easily demonstrated that it's not about real image quality in the final output, but rather the opposite.

My taste towards AA-filtered look has grown even stronger lately when I've started to make prints with an own printer, as I've seen even less reasons to have that pixel peep sharpness when I scale for printing, rather the opposite, in the rare event of having to scale up an aliased image will just make it look "digital".

It is an interesting question if removing the AA filter has a global effect, ie if you scale down the image so there's no resolution advantage, would you still see a difference (assuming we don't have visible moire in the subject). I've got the sense that there is a belief that there is indeed a "global advantage" in image quality of not having an AA filter, but I'm almost 100% positive that it's a myth (the only reason I would see is that AA filter would somehow disturb the efficiency of the color filters, but I haven't heard or read that there would be such a problem). I haven't done A/B testing though, there are not many cameras out there that allow that.

About that sharpening creates artificial detail, it depends on how you see it. The bayer+demosaicing process is softening in itself, and diffraction too. Counteracting those effects I would not say creates artificial details, ie you don't create any details that are not there you just enhance contrast. Aliasing creates false details for real though, in the form of small false colors and patterns around small details, which result in moire if the small details have a repeating pattern.

In landscape images moire occurs rarely, but false colors around small details occurs often. However with a 50 megapixel camera or so those details are so small so they generally become invisible in most publishing forms, and even if they are visible they're so tiny that the eye don't register that the color is wrong. So in practice it's certainly not a big issue for a landscape photographer like me. But I also know that the sligthly less sharp AA-filtered image would also not result in meaningful loss in sharpness, and then it for me feels better to rather trade (reasonably) accurate details for a visually reversible softening.

[torger]

Here's something that shows aliasing in a irregular landscape context. f/11 with 6um pixels, rendered with Phocus which is better than most raw converters at hiding false colors. Lightroom will not make this good result.

And no the tree branches are not magenta or cyan, they're dark brown with some snow on them.

However when the details are rendered small the eye won't see colors just luminance difference so you don't actually see that the color is wrong (maybe you don't even see it on the attached thumb, click it to see the enlargement), so as long as you have a high enough pixel count to render pixels tiny on your final output you're safe. So you generally don't need to worry, but I fail to see how rendering the wrong colors should be seen as a image quality feature. How well AA-filtered images can be sharpened is a different demonstration, but that will show that there's no meaningful loss in actual detail or look.

[EricWHiss]

Dear Sir,
I have had Imatest since it's inception. I have also published a lot of data using Imatest, which you have bin happy to criticise. You memory seems to be short.

The truth is your memory may be the shorter one .... Don't you remember that we agreed to ignore each other?  Since then I have happily overlooked a great number of your posts presumably many loaded with Imatest results you write about. But don't take it too personally as I've been quite short of time lately and spend much less time at LuLa now than ever.  And while I don't recollect critiquing any imatest results of yours,  I have in the past pointed out some flaws in your testing particularly some based on sample images of a building that looked impossibly soft for a Phase p45 back and Zeiss lens unless they were out of focus or possibly taken with a duff lens. It seems you took this as a personal attack?  The feeling of being attacked must be the only thing you do remember since you seem to relish in taking every chance to get in a little 'axe grinding' as you referred to it.  But why waste everyone else's time?  This shouldn't be personal, and if a person who professes to wish to learn, they should be the first one wanting to know if they have done something wrong in their tests.

Anyhow, I think you missed where I was going with the Imatest comment.  I was suggesting you make a little test to discover for yourself how little detail you would have actually lost stopping down to f/16 with the p45.  Was it worth keeping your lens at the f/stop that would result in aliasing just to avoid a little diffraction loss?  That's the question I was hoping you would answer for yourself.

I know with the 80mp back which has even smaller pixels than the p45, I just don't miss that much at f/16 even though I know diffraction starts at f/8.   And when its really important, I can get back some of that with various software sharpening tools that recover diffraction loss.     It seems from the comments of others, they either are experiencing aliasing to the degree you do, or are not bothered by it, or work around it.  I'm certainly in that camp.   Color aliasing just isn't a problem for me and in my work, I almost never think about it.  

For that reason I don't see color aliasing as enough motive by itself to drive the decision to make sensors with smaller pixels when the trade offs, as I explained already, are loss of DR, and less range of working DOF (for a given sensor format).   If a camera maker also shrinks the sensor size with the pixels then maybe it will work out for the DOF usability concerns, but even then the system will be reliant on expensive ultra fast lenses.      I'm more in favor of a larger sensor with 7um or bigger pixels, hopefully 6x6 or 6x7 in size or even bigger. 4x5 large format sensor?  Yes please!

Back to the topic of the Leica S,  I think the camera competes in a different market than the DSLR's even though it is like one.  The S offers a much nicer viewfinder, a bigger sensor, high build quality and a very nice set of lenses. I certainly wish Leica well. They've done an amazing job to develop a whole new camera system and lenses.   I'd like to have one myself, but so far have resisted because I already have too many cameras and not enough time.


  

[EricWHiss]

I don't think people generally "work around aliasing" (shaking the camera, shooting at f/22, shooting out of focus), they just ignore it. In most cases it's really subtle and clients won't notice. Only specialists on digital image quality will notice, unless you have some really bad moire (which indeed is quite rare).

Torger,
You could be right about the pixel peeping joy, psychology, or perhaps let's call it the ability to deeply crop an image.  I do remember quite a lot of discussions here on LuLa from working pros about how to deal with moire in fabrics and such when the standard pixel pitch for MFDB was 9um.  If you search the forum you'll bring it up.  Stopping down, moving subject to camera distance, and caprock filters were all discussed among other workarounds.

It will be interesting to see what resolution differences there are between the new Canon with and without the AA filter.  Probably someone is out there testing this right now as I type?

[ErikKaffehr]

Hi Eric,

In general there is no resolution difference between an OLP-filtered image and an OLP-filtered image. The resolution is essentially allways limited by the sensor, lens or diffraction. What the OLP-filter does is reducing MTF, that is contrast at the pixel pitch. So MTF is affected and it is affected very significantly.

If you photograph a Siemens star or even better Bart van der Wolf's test target, you will find that the lines converging start to turn around. That is when you pass the Nyquist limit.

I have never owned a camera that would not show these bending lines with a decent lens at medium apertures. The P45+ is the only camera I have that lacks OLP filtering.

The enclosed images show Barts's and Norman Koren's test targets shot with Sony Alpha 99, and P45+. The P45+ image has been developed in both LR5 and Capture One V8. I also added a shot from the A77 that has 3.8 micron pixels. All those images were shot at 1/50 magnification, so they illustrate the effect of pixel pitch and ignore format differences.

It is interesting to see the Capture One is quite sucessfull in hiding colour artefacts, specially on the Norman Koren target but it has the same type of monochrome artefacts as the image developed in LR5.

All the images have the bending lines, indicating that the sensor limits resolution, but the 3.8 micron pixel image is quite low on fine detail contrast.

Torger,
You could be right about the pixel peeping joy, psychology, or perhaps let's call it the ability to deeply crop an image.  I do remember quite a lot of discussions here on LuLa from working pros about how to deal with moire in fabrics and such when the standard pixel pitch for MFDB was 9um.  If you search the forum you'll bring it up.  Stopping down, moving subject to camera distance, and caprock filters were all discussed among other workarounds.

It will be interesting to see what resolution differences there are between the new Canon with and without the AA filter.  Probably someone is out there testing this right now as I type?

[ErikKaffehr]

The images shown here are screen dumps from Photoshop CS at 1:1. On the left, the same Alpha 77 (3.8 Micron) image as before but downscaled to 55% to match the size of the 6.8 micron P45+ image. The P45+ image is coming right out from C1 v8 with default setting. Some extra sharpening was added to the image on the left side and a small curve adjustment was done.

Check out the area marked in red. Text is still readable in the left image, but barely readable in the right.

An original version of that image is here: http://echophoto.dnsalias.net/ekr/Articles/Shoots/Aliasing/P45vsa77.png

Note, some of the artefacts in the left image are a consequence of the crude resizing in Photoshop CS. Lightroom makes a much better job at this.

Another experiment would be to resize the 6.8 micron image to the same size as the 3.8 micron, I have done that using bicubic in Adobe PSCS and enclose it as a second image.
Best regards
Erik

Hi Eric,

In general there is no resolution difference between an OLP-filtered image and an OLP-filtered image. The resolution is essentially allways limited by the sensor, lens or diffraction. What the OLP-filter does is reducing MTF, that is contrast at the pixel pitch. So MTF is affected and it is affected very significantly.

If you photograph a Siemens star or even better Bart van der Wolf's test target, you will find that the lines converging start to turn around. That is when you pass the Nyquist limit.

I have never owned a camera that would not show these bending lines with a decent lens at medium apertures. The P45+ is the only camera I have that lacks OLP filtering.

The enclosed images show Barts's and Norman Koren's test targets shot with Sony Alpha 99, and P45+. The P45+ image has been developed in both LR5 and Capture One V8. I also added a shot from the A77 that has 3.8 micron pixels. All those images were shot at 1/50 magnification, so they illustrate the effect of pixel pitch and ignore format differences.

It is interesting to see the Capture One is quite sucessfull in hiding colour artefacts, specially on the Norman Koren target but it has the same type of monochrome artefacts as the image developed in LR5.

All the images have the bending lines, indicating that the sensor limits resolution, but the 3.8 micron pixel image is quite low on fine detail contrast.

[torger]

Torger,
You could be right about the pixel peeping joy, psychology, or perhaps let's call it the ability to deeply crop an image.  I do remember quite a lot of discussions here on LuLa from working pros about how to deal with moire in fabrics and such when the standard pixel pitch for MFDB was 9um.  If you search the forum you'll bring it up.  Stopping down, moving subject to camera distance, and caprock filters were all discussed among other workarounds.

It will be interesting to see what resolution differences there are between the new Canon with and without the AA filter.  Probably someone is out there testing this right now as I type?

Fortunately aliasing with more recent cameras are much less than with the 9um microlens-free sensors, with lower resolution and no microlenses it gets worse.

It's not exactly hard to see it in an 80 megapixel file either though if you look at 100%, this one is riddled with aliasing:
http://en.wikipedia.org/wiki/Phase_One_%28company%29#mediaviewer/File:JacquelineMegaw.jpg
but there's fortunately no large areas of moire and then you get away with that the aliased parts will be so small in the final output that the eye does not detect the color errors. So today I think most just don't care. It's kind of ironic to increase the resolution with 400% to hide aliasing when you could reduce it with 1-2% with an AA filter and then get rid of it...

Probably we have to wait a while for the 5DS vs 5DS R comparison as the cameras are a few months away from public availability. There was some D800 vs D800E tests though, but mostly done by people that has already fallen for the common view that no AA filter equals better image quality. If you apply the same amount of sharpening on both and evaluate by looking at 100% pixel crops of course the filter-free will win, but if you instead evaluate with adapted sharpening and do it on upscaled prints and value color correctness the result will be different.

Bayer array is still by far the best color sensor technology, better color filters, better sensitvity, Foveon cannot match at all. But for Bayer to perform at its best it should have an AA filter and it should have a high pixel count. If Bayer outresolves the optical system the artifacts go away. We're not there yet by far though, so more megapixels please

Edit: attached crops from the IQ180 image, as you can see there are some demosaic failures on the vertical fur, and on the horizontal fur you can see that the colors of the hair is not correctly rendered even in this case when it's slightly out of focus (this happens as the image is shot with a large aperture so diffraction does not save the day, when looking at areas with the hair even a bit more out of focus you can see what the correct color should be). Again the thumb is scaled down, click to see 1:1. To me this is an image quality problem, you don't get return on all your 80 megapixels. It's a small problem as 80 megapixels means scaling down in most cases, but if I would really need the resolution I would be concerned about getting colors right also on the small details. Therefore it would still be better with an AA filter that would reduce resolution to say correspondingly 78 megapixels and get the colors mostly correct and without demosaicing failures.

MFD manufacturers can't introduce AA filters now though, I think it would be a sales flop. The view that being without AA filter is a feature rather than a bug is just too strongly anchored I think. People would pixelpeep and look only for sharpness, and not notice that aliasing has disappeared.

[eronald]

If Bayer outresolves the optical system the artifacts go away.

Torger,

 I think this was the funny quote of the day. And quite correct, of course.

 Unfortunately, the lens guys will keep telling customers that their cameras are better with sharper lenses, an obsolete thought, obviously, which dates from the time of film

 However for now, one can also make the artefacts go away easily with some vaseline on a filter

Edmund

[torger]

Attached a d800 vs d800e 100% crop, from files available from dpreview. D800E is to the left (no AA), D800 to the right (with AA).

The files have been sharpened taking input softness into account, meaning that there's more sharpening on the D800 file than the D800E. It's probably possible to make a better job than I just did, but it's a start.

Some observations; moire with demosaic failures are visible in the D800E image on the rails and some less visible issues on house wall up to the right, the diagonal lines are more jagged, there are more color errors on the small details. We also see that the D800 image is not entirely free from color errors -- AA filters are relatively weak. We also see that the microcontrast is a bit lower and you have a little "bleed" in high contrast transitions. You don't really see more details in the D800E image, but the tiniest details have higher contrast but also more errors.

It can also be noted that the exactly horizontal rails makes it easier for the demosaicer which make line extensions (usually in 8 directions).

Which image represents the better image quality? It's a matter of taste, I think the AA-filtered version is better, especially when making real outputs such as prints, less jagged, less errors. However, I can't deny that with 1:1 pixel matching the trees appears a bit more detailed in the D800E for example, and the parts of the image that don't have aliasing and demosaic issues then the D800E looks a little bit better as there's less bleed and higher micro contrast. I don't think that advantage is real in a print though, I would worry more about the occasional small blotches of demosaic failures and the overall higher error rate.

[eronald]

Torger -

 if you can afford to downscale quite a bit, the aa filtered version will be cleaner, as Erik demonstrated.
 but at max rez or near to it, I feel the non-filtered version may sharpen/upscale better for a crisper if artefacted print.
 in a natural  landscape image you won't find all these straight line repeats so the artefacts won't be too much of an issue, while crispness will be noted.

e.

Attached a d800 vs d800e 100% crop, from files available from dpreview. D800E is to the left (no AA), D800 to the right (with AA).

The files have been sharpened taking input softness into account, meaning that there's more sharpening on the D800 file than the D800E. It's probably possible to make a better job than I just did, but it's a start.

Some observations; moire with demosaic failures are visible in the D800E image on the rails and some less visible issues on house wall up to the right, the diagonal lines are more jagged, there are more color errors on the small details. We also see that the D800 image is not entirely free from color errors -- AA filters are relatively weak. We also see that the microcontrast is a bit lower and you have a little "bleed" in high contrast transitions. You don't really see more details in the D800E image, but the tiniest details have higher contrast but also more errors.

It can also be noted that the exactly horizontal rails makes it easier for the demosaicer which make line extensions (usually in 8 directions).

Which image represents the better image quality? It's a matter of taste, I think the AA-filtered version is better, especially when making real outputs such as prints, less jagged, less errors. However, I can't deny that with 1:1 pixel matching the trees appears a bit more detailed in the D800E for example, and the parts of the image that don't have aliasing and demosaic issues then the D800E looks a little bit better as there's less bleed and higher micro contrast. I don't think that advantage is real in a print though, I would worry more about the occasional small blotches of demosaic failures and the overall higher error rate.

[torger]

if you can afford to downscale quite a bit, the aa filtered version will be cleaner, as Erik demonstrated.
 but at max rez or near to it, I feel the non-filtered version may sharpen/upscale better for a crisper if artefacted print.
 in a natural  landscape image you won't find all these straight line repeats so the artefacts won't be too much of an issue, while crispness will be noted.

I guess one have to make more testing with nosing real prints. I don't think the crispiness will be much noted as I think this effect is much enhanced on screen due to the one-to-one pixel match, but there could be some ppi range where crispiness may come through more. A guess would be that it would be in the 200 - 100 ppi range, and below 100 ppi jaggies and stuff will be problematic, and higher than 200 ppi the difference is just too small. This with inkjets which has rather fuzzy "rasters". In offset printing there is an "artificial" crispiness created by that the raster is very sharp at a smaller detail level than the underlying image, I like that look actually it makes the image to appear have detail into infinity without any softness at all.

One thing is for sure though, the AA-filtered version is more robust for scaling to any size.

When printing you'd do some additional sharpening tricks too, I'm a beginner at that though but it seems like some haloing can be an advantage to counteract print smearing, at least that was the case with C-prints.

In landscape aliasing is indeed much less an issue than in architecture. Moiré that I have myself come across in landscape has happened in rippled water and in sand or gravel, but that is rare. Aliasing in the form of false colors around small details happen often though, narrow branches against a bright background is a typical example, or now in a snowy landscape where you have high contrast edges between snow and dark things below it. As said earlier those details are generally so small that in a typical viewing scenario the eye would not register color. Maybe one could argue that the eye would register the increased micro-contrast but not the color errors. The eye does have higher luminance resolution than color resolution. It would be an interesting experiment indeed!

If one would find that the crispiness of the 36 megapixel D800E can be appreciated in a real print it would be interesting to see if an AA-filtered 40 megapixel image of the same FOV would appear even more crisp. Unfortunately there are no cameras to make such test, but I think it could show that the slightly higher resolution AA-filtered camera would be even more crisp, so what you in practice get when getting rid of the AA filter is just getting a very slight virtual increase in megapixels, traded for potential image quality issues.

From a "philosophical" perspective I don't find it hard to prefer the AA-filtered version simply because it makes a more correct image at a very minor cost, but I also see that you can come to a different conclusion. It's not black and white.

As a software engineer I've already started thinking that you could do better aliasing suppression than what's available today too, you could have some sort of post-processing pass where you detect high contrast narrow details where you know aliasing would exist and simply remove magenta/cyan there. Possibly super-resolution techniques could be used along edges to figure out the likely non-aliased color. Demosaicers today are more focused on speed than quality, there should be more to do there. I also think more can be done in upscaling, making a crossover between bicubic and fractal/vector to avoid jaggies while keeping a natural look. It does not really matter how much I like AA-filtered images from an image quality perspective, AA filters is not coming back to medium format and in 135 I'm quite sure 5DS will be the last high res camera from Canon with AA-filter, because people will choose the 5DS R based on pixel peep joy alone.

[paratom]

Wow, I find your remarks remarkable.  I also prefer Leica's M colors to Fuji's but found Leica's less saturated than Fuji's.  I now use both X-Pro1 and X-T1 and often turn down saturation.  I often find Fuji's colors too velvia like (I disliked velvia).  Interesting, how our perceptions differ.

I also liked the colors Zeiss lenses on my Canon 5D gave me more appealing (cooler) than from Canon lenses.  A shame that manual focusing with the 5D is such a pain.

Mmmh, maybe different raw converter or processing? I was talking about x-pro 1 in LR vs Leica M in LR.