Pages: [1] 2 3 ... 7   Go Down

Author Topic: AA-filtering CCD and CMOS  (Read 38553 times)

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
AA-filtering CCD and CMOS
« on: October 02, 2009, 01:09:34 am »

Hi,

I have scanned trough some articles regarding MTF and "fill factors". It seems that there is a lot of research going on about MTF of sensors at the subpixel level. One issue that came to my mind is that CCD sensor cells approach 100% fill factor (according to some of the articles) while CMOS sensors have a much smaller "fill factor" and the actual sensel area is quite irregular, usually "L-shaped".

As CCD-s seem to have larger fill factor, it seems that they would be in less need of AA-filtering (or OLP-filtering to use another word). This may explain, in part, while MFDBs and Leica can do without AA-filter.

Partly, because it's obvious that both MFDBs and Leica M9 can have "color moiré" in certain cases.

Just "Google" for "fill factor and MTF" and you will find a lot of articles....

Best regards
Erik


Logged
Erik Kaffehr
 

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13983
    • http://www.flickr.com/photos/bernardlanguillier/sets/
AA-filtering CCD and CMOS
« Reply #1 on: October 02, 2009, 01:19:44 am »

Quote from: ErikKaffehr
I have scanned trough some articles regarding MTF and "fill factors". It seems that there is a lot of research going on about MTF of sensors at the subpixel level. One issue that came to my mind is that CCD sensor cells approach 100% fill factor (according to some of the articles) while CMOS sensors have a much smaller "fill factor" and the actual sensel area is quite irregular, usually "L-shaped".

Beyond these totally relevant technical factors, the biggest difference between CCD and CMOS is probably the fact that Japanese companies all use CMOS with AA filters while non Japanese companies all use CCD without AA filters.

It would be interesting to think about the reasons why these different technological options were selected by different engineering cultures, but I am more interested today in the downstream part of the process, the communication between brands and reviewers.

It is obvious that Western reviewers like Michael have much closer contacts with Western brands using CCDs without AA filter than they have with Japanese brands using CMOS with AA filters.

It seems fair to say that the promoters of CMOS with AA filters have not invested as much effort in communicating to Western reviewers about their motivations. We are basically only hearing one side of the story.

Cheers,
Bernard

Slough

  • Guest
AA-filtering CCD and CMOS
« Reply #2 on: October 02, 2009, 03:23:17 am »

Another area that gets little attention is gamut, which surely is due to the choice of colour filters. I suspect the Japanese manufacturers attend more to things that look good in specs and tets such as high ISO performance, and pixel count i.e. the things that dpreview pixie peepers focus on.
Logged

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13983
    • http://www.flickr.com/photos/bernardlanguillier/sets/
AA-filtering CCD and CMOS
« Reply #3 on: October 02, 2009, 04:28:22 am »

Quote from: Slough
Another area that gets little attention is gamut, which surely is due to the choice of colour filters. I suspect the Japanese manufacturers attend more to things that look good in specs and tets such as high ISO performance, and pixel count i.e. the things that dpreview pixie peepers focus on.

Well... they are not alone. I remember being laughed at big time when I claimed last year on this very forum that true RGB devices like Betterlight backs delivered significantly better colors...

Cheers,
Bernard

georgl

  • Full Member
  • ***
  • Offline Offline
  • Posts: 140
AA-filtering CCD and CMOS
« Reply #4 on: October 02, 2009, 04:56:34 am »

The fill-rate (amount of light-sensitive area) is IQ-wise one of the most important technical aspects of sensors. The higher the fill-factor, more photons hit the (larger) light-sensitive surface and more photons can hit it before saturation sets in (important for DR). Of course the analogue sensor signal then has to be amplified and converted in digital information - but what's lost in the sensor, cannot be restored.

Full-frame CCDs need special manufacturing sites (they're expensive), they're slow, they consume much energy and they're stupid (they're just a large array of photodiodes). But they have one big advantage: fill-rate, it's still not 100% but close.
So they are able to gather more information, create a high-quality and still pretty much unprocessed (no filtering etc.) signal. That's why they are used for prefessional digital cameras.

CMOS-based systems have come a long way, but their fill-rate is still quite low but they're capable of very sophisticated image-processing to reduce noise (and they have to, especially certain kinds of pattern noise due to the specific amplification process of CMOS), that's why high-iso-files from these cameras appear quite clean but also tend to be more artificial.
When comparing noise, we have to compare processed (because CMOS-files are always filtered internally) RAW-files regarding noise AND detail. Regarding CMOS and CCD we then have to care about the same pixel-pitch, usage of microlenses and the age of the architecture. The common 6.8µm-CCDs are from 2004! We will have such a rare opportunity to compare "CMOS vs. CCD" again with the appearence of the S2 - it has microlenses (the other 6µm-CCD-systems don't have those) and is comparable to the 20+MP-DSLRs and when the engineers didn't mess up the processing (conversion...) the noise/detail will look different but won't be worse than with CMOS-based systems.

CMOS is the future, the new EVIL-systems need their capability to implement live-view, even for professional camera systems. But the approach will be different, when not using AA-filters, reduce internal processing to minimum and similar HQ-color-filters as in current CCDs, their "look" won't be much different than current CMOS - if there will be a difference at all!
CMOS or CCD-sensors itself don't have a "look", they're just electronic devices to convert light into electric information! The size of the sensels, the microlenses, the architecture etc. don't even affect sharpness/MTF because these things are all affecting individual pixels (just like sharpness is no longer an issue with TFT-displays instead of CRTs). Only the AA-filter affects "sharpness" sensor-wise (given the same pixel-pitch/size)

AA-filters are needed to reduce alaising artifacts "out of the sensor" and have nothing to do with the choice of CCD vs. CMOS. That's what the Japanese DSLRs are designed for: professional press-photography in the upper end and amateurs on the lower end. They both need their images fast without post-processing, even if they have to compromise IQ. MFDB-files are processed carefully in post.
CCDs can be used without microlenses, making them more sensitive to oblique light rays and avoid aberrations - important for technical cameras. But you loose about one stop of sensitivity, because microlenses try to compensate the low fill-rate and focus light only on light-sensitive areas of the sensor. That's why they are more important with CMOS-designs, their "loss in sensitivity" due to the lack of microlenses would be much bigger.
« Last Edit: October 02, 2009, 04:08:58 pm by georgl »
Logged

Daniel Browning

  • Full Member
  • ***
  • Offline Offline
  • Posts: 142
AA-filtering CCD and CMOS
« Reply #5 on: October 02, 2009, 02:31:44 pm »

Quote from: ErikKaffehr
As CCD-s seem to have larger fill factor, it seems that they would be in less need of AA-filtering (or OLP-filtering to use another word). This may explain, in part, while MFDBs and Leica can do without AA-filter.

Yes, the optical fill factor does relate to how bad the aliasing will be without an OLPF; however, CCD has not had an advantage in optical fill factor for many, many years thanks to the invention of microlenses. Digicams got up to 100% optical fill factor ~4 years ago, but it was only two years ago that 36x24mm DLSRs finally hit that last few percent to reach 100.

So no, this has not been a reason for the difference in AA-filtering for a long time.
Logged
--Daniel

aaykay

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 359
AA-filtering CCD and CMOS
« Reply #6 on: October 02, 2009, 05:55:21 pm »

Quote from: BernardLanguillier
Beyond these totally relevant technical factors, the biggest difference between CCD and CMOS is probably the fact that Japanese companies all use CMOS with AA filters while non Japanese companies all use CCD without AA filters.

And then the fact that there are Japanese sourced CCDs with AA filters....   .....proving that the usage of AA filters have nothing to do with CCD or CMOS but a deliberate design choice by the camera (not sensor) manufacturers.

Logged

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13983
    • http://www.flickr.com/photos/bernardlanguillier/sets/
AA-filtering CCD and CMOS
« Reply #7 on: October 02, 2009, 06:36:38 pm »

Quote from: aaykay
And then the fact that there are Japanese sourced CCDs with AA filters....   .....proving that the usage of AA filters have nothing to do with CCD or CMOS but a deliberate design choice by the camera (not sensor) manufacturers.

Yes, that is indeed correct if you factor in compact cameras. Aren't all Japanese reflex CMOS nowadays?
 
Regards,
Bernard

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
AA-filtering CCD and CMOS
« Reply #8 on: October 02, 2009, 06:40:49 pm »

Beranrd,

There are some CCD based designs around.

I suggest that you check out this: http://www.imx.nl/photo/leica/camera/page157/page157.html

Best regards
Erik

Quote from: BernardLanguillier
Yes, that is indeed correct if you factor in compact cameras. Aren't all Japanese reflex CMOS nowadays?
 
Regards,
Bernard
Logged
Erik Kaffehr
 

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13983
    • http://www.flickr.com/photos/bernardlanguillier/sets/
AA-filtering CCD and CMOS
« Reply #9 on: October 02, 2009, 06:58:21 pm »

Quote from: ErikKaffehr
Beranrd,

There are some CCD based designs around.

I suggest that you check out this: http://www.imx.nl/photo/leica/camera/page157/page157.html

I was speaking about Japanese DSLRs Erik.

Cheers,
Bernard

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
AA-filtering CCD and CMOS
« Reply #10 on: October 03, 2009, 12:46:06 am »

Sorry Bernard!

The two comments were not related. Some of Sony's low end DSLRs are CCD-based, there may be some others.

The Erwin Puts article is interesting because he compares the optical performance of the Nikon D3X with the M9. In short:

- D3X comes out on top
- M9 is very good
- But M9 has also Moiré

That helicopter shoot of yours is very sharp, BTW, you have a recepie for sharpening?

Best regards
Erik

Quote from: BernardLanguillier
I was speaking about Japanese DSLRs Erik.

Cheers,
Bernard
Logged
Erik Kaffehr
 

TheSuede

  • Newbie
  • *
  • Offline Offline
  • Posts: 16
AA-filtering CCD and CMOS
« Reply #11 on: October 03, 2009, 09:22:51 am »

Hi all... :-)

I'd actually say that working without an AA-filter only makes your pictures more detailed (and sharper) in SOME cases. As Fraunhofer (and Kodak, and Dalsa, and several independant studies I've seen) have found, more complex targets with smaller detail or colour-driven detail more often than not get a lower dE ("perfect" original to raw-interpolated picture, pixel-per-pixel value deviation) when you shoot (or simulate) a sensor with an AA-filter and sharpen it right. This is quite logical if you know something about sampling theory and the Bayer CFA principle. Any one pixel HAS to have some relevant and statistically conformant relationship to the surrounding pixels if the interpolation of the two missing colours per pixel is to have any chance at all to do the right estimation. Most studies use the standard Kodak picture targets, software simulated as CFA filtered raws.

The absolutely best results in "mixed" real world picture targets seems to be a controlled 8% light spread in the X- and Y-axis (none in the 45º axises), this gives the raw-interpolation engine maximum detail to work with, with reasonably good relevance to the reality in front of the lens. What is interpreted as "detail" and "sharpness" in an AA-filter-less sensor package is actually mostly interpolation noise and high-frequency detail miscalculations. It has very little to do with real detail and sharpness - except in cases as simple as the Imatest slanted edge measurement - which is a very simple "black square on white background, how sharp is the outline"-test. No detail, no colour.

Easy target, no fine detail or fine colour detail - AA-less is better. ALL other cases - Controlled AA-filter with the right strength is better. Unless you're truly comfortable with the fact that the picture isn't conformant with reality. In Fraunhofer's study, the impression of "more detail" in the AA-filter less sensor could be totally negated by adding a very fine pitch noise to the sharpened AA-filtered version - then the filtered version was perceived as "more detailed", and STILL had better conformance with reality - pixel for pixel as measured by standard deviation from the "perfect" original.

I'm also very surprised that such a seasoned and knowledgeable person as Erwin gives the Leica 250% oversharpening and forgets to scale the resolutions evenly in the comparison graphs... :-) Note that the D3x resolutions are measured out to "2500", and the M9 uses the same graph width, but is stretched to make "1800" the maximum value out right in the graph.... Do a mental correction for the scale differences before you compare the graphs - then the story looks even more "definitely not in favour" of the M9... :-)


Quite a rant for a first post, but I was actually quite "not impressed" with Erwin's latest study - he is usually a VERY dependable and neutral person. Even if his entire salary comes from the Leica PR-department... :-)
Logged

aaykay

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 359
AA-filtering CCD and CMOS
« Reply #12 on: October 03, 2009, 09:49:49 am »

Quote from: BernardLanguillier
I was speaking about Japanese DSLRs Erik.

Cheers,
Bernard

And I think that is precisely what is being talked about too.  DSLRs.  The  current Sony A300/350/330/380 along with some of the lower-end Nikon models (D40/D40X/D60), the previous D200 and all of the Pentax models prior to the recent ones employing the Samsung CMOS chip, used CCDs with AA filters.  

In all of their semi-pro and above models, Sony has completely transitioned into CMOS and only uses CCD in their lowest-end consumer models.  Same story with Nikon, where in the past only the D2X came with CMOS but now all of their semi-pro and Pro models have moved away from CCD into CMOS.  Pentax too have transitioned from CCD into CMOS but do retain some CCDs in their lowest-end consumer models.
« Last Edit: October 03, 2009, 09:53:26 am by aaykay »
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
AA-filtering CCD and CMOS
« Reply #13 on: October 03, 2009, 09:59:58 am »

Hi,

It may be helpful if you could give links to studies you refer to. Even without references I'd say that your comments put the AA-filtering issue in perspective. Regarding Erwin Puts's articles I have not noticed the 250% oversharpening. That said I often feel that data presented by Erwin Puts is not always easy to grasp or compare but I get the impression that he puts a lot of efforts into his tests.

To me it seems odd that DSLR makers would spend resources on quite expensive filters, something that would reduce image quality. It simply doesn't make any sense. Therefore I assume that they are employed for good reason.

Best regards
Erik

Quote from: TheSuede
Hi all... :-)

I'd actually say that working without an AA-filter only makes your pictures more detailed (and sharper) in SOME cases. As Fraunhofer (and Kodak, and Dalsa, and several independant studies I've seen) have found, more complex targets with smaller detail or colour-driven detail more often than not get a lower dE ("perfect" original to raw-interpolated picture, pixel-per-pixel value deviation) when you shoot (or simulate) a sensor with an AA-filter and sharpen it right. This is quite logical if you know something about sampling theory and the Bayer CFA principle. Any one pixel HAS to have some relevant and statistically conformant relationship to the surrounding pixels if the interpolation of the two missing colours per pixel is to have any chance at all to do the right estimation. Most studies use the standard Kodak picture targets, software simulated as CFA filtered raws.

The absolutely best results in "mixed" real world picture targets seems to be a controlled 8% light spread in the X- and Y-axis (none in the 45º axises), this gives the raw-interpolation engine maximum detail to work with, with reasonably good relevance to the reality in front of the lens. What is interpreted as "detail" and "sharpness" in an AA-filter-less sensor package is actually mostly interpolation noise and high-frequency detail miscalculations. It has very little to do with real detail and sharpness - except in cases as simple as the Imatest slanted edge measurement - which is a very simple "black square on white background, how sharp is the outline"-test. No detail, no colour.

Easy target, no fine detail or fine colour detail - AA-less is better. ALL other cases - Controlled AA-filter with the right strength is better. Unless you're truly comfortable with the fact that the picture isn't conformant with reality. In Fraunhofer's study, the impression of "more detail" in the AA-filter less sensor could be totally negated by adding a very fine pitch noise to the sharpened AA-filtered version - then the filtered version was perceived as "more detailed", and STILL had better conformance with reality - pixel for pixel as measured by standard deviation from the "perfect" original.

I'm also very surprised that such a seasoned and knowledgeable person as Erwin gives the Leica 250% oversharpening and forgets to scale the resolutions evenly in the comparison graphs... :-) Note that the D3x resolutions are measured out to "2500", and the M9 uses the same graph width, but is stretched to make "1800" the maximum value out right in the graph.... Do a mental correction for the scale differences before you compare the graphs - then the story looks even more "definitely not in favour" of the M9... :-)


Quite a rant for a first post, but I was actually quite "not impressed" with Erwin's latest study - he is usually a VERY dependable and neutral person. Even if his entire salary comes from the Leica PR-department... :-)
« Last Edit: October 03, 2009, 10:05:49 am by ErikKaffehr »
Logged
Erik Kaffehr
 

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
AA-filtering CCD and CMOS
« Reply #14 on: October 03, 2009, 10:04:14 am »

Yes,

Just would point out that normally a double layer of OLP are employed, one vertical and one horizontal. Some of the Pentax models seem to have only one OLP (either vertical or horizontal).

Best regards
Erik


Quote from: aaykay
And I think that is precisely what is being talked about too.  DSLRs.  The  current Sony A300/350/330/380 along with some of the lower-end Nikon models (D40/D40X/D60), the previous D200 and all of the Pentax models prior to the recent ones employing the Samsung CMOS chip, used CCDs with AA filters.  

In all of their semi-pro and above models, Sony has completely transitioned into CMOS and only uses CCD in their lowest-end consumer models.  Same story with Nikon, where in the past only the D2X came with CMOS but now all of their semi-pro and Pro models have moved away from CCD into CMOS.  Pentax too have transitioned from CCD into CMOS but do retain some CCDs in their lowest-end consumer models.
Logged
Erik Kaffehr
 

georgl

  • Full Member
  • ***
  • Offline Offline
  • Posts: 140
AA-filtering CCD and CMOS
« Reply #15 on: October 05, 2009, 01:04:37 pm »

@ErikKaffehr
AA-filters are needed if you have to avoid moire at any cost. Press-photographers usually don't have the time to post-process the images to remove moire - they rather live the loss of fine detail/contrast.

Mr. Puts has made some interesting articles about lenses and optical design, this was also his last direct connection to Leica (a brochure in 2005?). But his articles about digital photography are quite often misleading and sometimes even wrong.

@Daniel Browning
The "look" of moire has definitely changed with the various sensor-generations (with increasing fill-rate) but alaising itself appears nevertheless and never seemed to influenced the choice of AA-filter vs. unfiltered!?

@TheSuede
The aspect of bayer-interpolation is quite interesting, but I'm not familiar with these studies.

But the basics of alaising and AA-filters remain unchanged, moire doesn't destroy detail and AA-filters don't preserve it.
« Last Edit: October 06, 2009, 04:24:37 am by georgl »
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
AA-filtering CCD and CMOS
« Reply #16 on: October 05, 2009, 02:01:56 pm »

Quote from: BernardLanguillier
Aren't all Japanese reflex CMOS nowadays?
No. Not the brand new Nikon D3000 for example, or this year's Sony A220, A330 and A380, or the Pentax K2000 (K-m) and K200D. The Sony 10MP CCD is the main one left is use though; five of the above six use it. I suspect that Sony is squeezing the last value out of its "paid for" DSLR CCD production lines.

Anyway, over the years there have been a great many Japanese DSLR's with CCD sensors, all with AA filters, so I see no connection between AA filter and CCD/CMOS.

It might have more to do with the high end customers (MF, Leica) leaning towards optimizing sharpness and such, and being willing to do some occasional PP on moiré to get the best IQ, whereas the mainstream is more likely to sacrifice the last few percent of sharpness for more consistently usable "out of the box" results.
« Last Edit: October 05, 2009, 02:04:15 pm by BJL »
Logged

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13983
    • http://www.flickr.com/photos/bernardlanguillier/sets/
AA-filtering CCD and CMOS
« Reply #17 on: October 05, 2009, 07:09:30 pm »

Quote from: ErikKaffehr
It may be helpful if you could give links to studies you refer to. Even without references I'd say that your comments put the AA-filtering issue in perspective.

Which is exactly what I was refering to above.

The opinion of the back/M8/M9 owners has mostly been formed based on information coming from companies doing AA filter less devices. Once you start to listen to both sides of the story, it becomes a lot less clear cut... and as we are seeing with this contribution, there are many people around who believe that the lack of AA filter has little value if any in actual detail rendering.

Cheers,
Bernard

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13983
    • http://www.flickr.com/photos/bernardlanguillier/sets/
AA-filtering CCD and CMOS
« Reply #18 on: October 05, 2009, 07:24:01 pm »

Quote from: georgl
@TheSuede
The aspect of bayer-interpolation is quite interesting, but I'm not familiar with these studies.

But the basics of alaising and AA-filters remain unchanged, moire doesn't destroy detail and AA-filters don't preserve it.

What is being said here is different. According to this contribution, AA filter less devices are intrinsically unable to capture the reality of a scene.

We have gotten used to the smooth fonts we see on our screens but sometimes fail to realize that the disapearance of stair cases on our display is the result of a voluntary blurring of their edges... another example where anti-aliasing is required to make a shape look the way it actually is beyond the limited spatial resolution of the digital information.

All in all, the over-simplications that are often proposed by one party to sell a technology are typically poor basis for judgement. The only option is to listen to ask the question to both parties and listen to both over-simplifications.

Another way is to look at 100% crops and make up one's mind based on the actual results delivered by the technology. Every time I do so, I end up preferring the way the correctly sharpened AA filter image looks.

Cheers,
Bernard

telyt

  • Guest
AA-filtering CCD and CMOS
« Reply #19 on: October 07, 2009, 07:45:27 pm »

Quote from: BernardLanguillier
The opinion of the back/M8/M9 owners has mostly been formed based on information coming from companies doing AA filter less devices.

I'm not sure how you determined this.  The M8/M9/DMR owners I know of who prefer the AA filter-less approach formed their opinions by comparing output from the AA filter-less camera to their other, AA-filtered, cameras.
Logged
Pages: [1] 2 3 ... 7   Go Up