CCD or CMOS, interesting experiment by David Farkas at red dot forum

[ErikKaffehr]

Hi,

There has been a lot of discussion about CCD and CMOS. David Farkas made an experiment shooting the M9 and M(240)

The first part, he tries to match M(240) images to M9 under a set of conditions that favor the M9, it is here:
http://www.reddotforum.com/content/2015/02/the-great-debate-ccd-vs-cmos-part-1/

The next part he presents alternative images to see if viewers can identify the camera they have been shot with:
http://www.reddotforum.com/content/2015/02/the-great-debate-ccd-vs-cmos-part-2/

The third instalment will discuss the findings.

Best regards
Erik

[wmchauncey]

IMHO, evaluating image IQ based on internet images is a complete exercise in futility.     

[ErikKaffehr]

Hi,

I definitively agree regarding image quality, but it has been suggested that they would give very different colour, and that should be visible also at web size.

Best regards
Erik

IMHO, evaluating image IQ based on internet images is a complete exercise in futility.     

[Paul2660]

I test I have done many times, and to me the results are always close.  Where I still see a better look/feel with CCD tends not to be colors, but in areas like details on rocks, (lichen) finer tree limbs off in the distance etc.  Whereas the CMOS always has a better shadow recovery.

Paul

[Phil Indeblanc]

So in essence you're saying CCD IS better for fine detail photographic needs.

[uheck]

David Farkas is a supernice guy.
Super humble.Soft spoken.Educated.
But NEVER neutral.

A Leica PR spokesperson.

[ErikKaffehr]

Yes,

I have noticed both. But in this case he compares Leica M9 to Leica M typ 240. Also, he just posts samples, even if those samples are processed. So it is a blind test.

Anyway, I would guess it is a good indication if there is an obvious difference between the two.

Best regards
Erik

[bjanes]

IMHO, evaluating image IQ based on internet images is a complete exercise in futility.     

Not only that, but there are many variables in the equation other than CCD vs CMOS. Older Kodak CCD sensor vs newer CMOSIS designed sensor made in a French fab. Different optical stacks, different Bayer filters, different microlenses, different electronics, and possibly different post processing. All CMOS sensors are not alike as shown by DXO's comparison of a Nikon vs Canon sensor. They describe some of the variables within the same category of sensor (CMOS in this case).

Bill

[Paul2660]

So in essence you're saying CCD IS better for fine detail photographic needs.

Phil

To be fair to CMOS, the largest I can compare is a D810 to say a 80MP CCD.  The D810 and D800 with a good lens system can get very close if not the same degree of details.  When I wrote that first comment, I was not considering that I was uprezing the D800 files to the size of the native 60 and 80MP shots.  Uprezing always reduces finer details to my eyes. 

A stitched D810 can do very well indeed. 

What you tend to see is just a bit of difference in the rendering of the smaller details, CMOS vs CCD.  I am sure that the new Canon 50Mp sensors with some of the new lenses that Canon has come out with will offer some excellent results. 

Paul

[ErikKaffehr]

Hi Bill,

Absolutely agree with every word. But…

It is a wide spread conception with many photographers that CCDs have some magic property producing files that are superior. By that assumption viewers should be able to easily pick one type of camera over the other. If they cannot tell them apart it sort of indicates that the conception of CCD having superior colour to CMOS is a misconception.

Best regards
Erik

Not only that, but there are many variables in the equation other than CCD vs CMOS. Older Kodak CCD sensor vs newer CMOSIS designed sensor made in a French fab. Different optical stacks, different Bayer filters, different microlenses, different electronics, and possibly different post processing. All CMOS sensors are not alike as shown by DXO's comparison of a Nikon vs Canon sensor. They describe some of the variables within the same category of sensor (CMOS in this case).

Bill

[JV]

Hi Bill,

Absolutely agree with every word. But…

It is a wide spread conception with many photographers that CCDs have some magic property producing files that are superior. By that assumption viewers should be able to easily pick one type of camera over the other. If they cannot tell them apart it sort of indicates that the conception of CCD having superior colour to CMOS is a misconception.

Best regards
Erik

Erik,

It has nothing to do with being superior.  Some people simply tend to prefer one look over another look. 

Likewise some people prefer the rendering of one lens, others the rendering of another lens.  It does not necessarily mean that one is superior over the other either. 

Although I have a lot of respect for David Farkas and the people from the Leica Miami store (I bought my S2 there) I am not sure what he is trying to prove with this test...

I personally have no doubt in my mind that through post you can make a CMOS file look like a CCD file.  IMO that only proves that you have reasonable to good post skills...

But that's not the point...

The point is that some people still prefer the CCD files, Leica still sells about 350 M-E cameras (essentially an M9) per month and recently there was an online petition to keep the CCD Leica alive...

On eBay the M9 is still trending at $3,300.  That's pretty good for a 6 year old digital camera...

Best, Joris.

[Chris Livsey]

On eBay the M9 is still trending at $3,300.  That's pretty good for a 6 year old digital camera...
Best, Joris.

Especially one that is odds on to need its sensor replacing. (regularly apparently until a "new" version is produced)
Disclaimer: Yes, at the makers cost but still 

[Abe R. Ration]

Eric Fossum about the colours in this context: http://www.dpreview.com/forums/post/54028441

The context below is the actual CMOS/CCD devices in use in the relevant products.

The image sensor is nothing but a device which samples the image which the lens draws. If the optical stack on top of the sensor is similar, there is little to choose between the light of the image which hits the sensor surface and the optical stack has little to do with the underlaying sampling device technology. When it comes to sampling this light the modern CMOS sensors tend to do (much) more efficient job than the CCDs used in Leicas and the medium format cameras, and the lower the exposure, the greater the difference.

The preference of some to CCD over CMOS is not based on superiority of the former as it's long been inferior if we consider evidence objectively (e.g. DxOMark). They don't offer "better colour accuracy" or "sharper details", but lower signal to noise ratio (SNR) which, if anything, causes worse colour accuracy and less crisp details. Of course if one system has more pixels than the other more details may be sampled or larger sensor which improves SNR, but that's beside the point of CCD vs. CMOS.

The arguments on "different look" of CCD vs. CMOS is in my opinion somewhat absurd, as the "look" is defined by the processing of the signal (e.g. Lightroom, Capture One, Photoshop,...), not the image sensor technology as long as we're well withing the comfort zone of both technologies when it comes to SNR. Beyond that CMOS can do what CCD, but not the other way around.

Some people may well prefer CCD sensor in their cameras, but it's just psycology.

[Bart_van_der_Wolf]

Eric Fossum about the colours in this context: http://www.dpreview.com/forums/post/54028441

The context below is the actual CMOS/CCD devices in use in the relevant products.

The image sensor is nothing but a device which samples the image which the lens draws. If the optical stack on top of the sensor is similar, there is little to choose between the light of the image which hits the sensor surface and the optical stack has little to do with the underlaying sampling device technology. When it comes to sampling this light the modern CMOS sensors tend to do (much) more efficient job than the CCDs used in Leicas and the medium format cameras, and the lower the exposure, the greater the difference.

The preference of some to CCD over CMOS is not based on superiority of the former as it's long been inferior if we consider evidence objectively (e.g. DxOMark). They don't offer "better colour accuracy" or "sharper details", but lower signal to noise ratio (SNR) which, if anything, causes worse colour accuracy and less crisp details. Of course if one system has more pixels than the other more details may be sampled or larger sensor which improves SNR, but that's beside the point of CCD vs. CMOS.

The arguments on "different look" of CCD vs. CMOS is in my opinion somewhat absurd, as the "look" is defined by the processing of the signal (e.g. Lightroom, Capture One, Photoshop,...), not the image sensor technology as long as we're well withing the comfort zone of both technologies when it comes to SNR. Beyond that CMOS can do what CCD, but not the other way around.

Some people may well prefer CCD sensor in their cameras, but it's just psycology.

+1

It has been explained before, photons do not care about the sensor that records them. It's the sensor dimensions/filtration and post-processing that defines the 'look'. CMOS devices just produce lower read noise, and hence a higher S/N ratio which benefits low light exposures. CMOS devices also have a few other benefits like lower power consumption (good for the battery requirements) and thus they operate cooler, which helps to reduce dark current from spoiling long exposures.

Cheers,
Bart

[logeeker]

I can't really believe that there are still some people who argue about this issue in 2015. Without any doubt CMOS is absolutely much better than any CCD.

[JV]

It's the sensor dimensions/filtration and post-processing that defines the 'look'.

So basically what you are saying is that if people e.g. prefer M9 images over M240 images it is because of different processing rather than because of inherent sensor characteristics?

[ErikKaffehr]

It is a bit more complex:

The sensor just captures photons and converts to electron charges. Color is applied by a color filter grid array in front of the sensor. That is made by a chemical process and may differ between vendors. But, it has nothing to do with CCD or CMOS.

After exposure, raw processing comes into play. The image is now just numbers, but some numbers belong to red, blue or green pixels. Each position has one colour and lacks the other two. The missing to colours are interpolated from surrounding pixels. That is called demosaicing.

After demosaic we have assigned three color responses to each and every pixels. Now, the raw converter will use a "compromise matrix" to convert the color responses into an internal representation.  Different profiles are thane applied to that internal colour response.

So there is quite a lot of math generating the final colours in each pixel.

Best regards
Erik

So basically what you are saying is that if people e.g. prefer M9 images over M240 images it is because of different processing rather than because of inherent sensor characteristics?

[Bart_van_der_Wolf]

So basically what you are saying is that if people e.g. prefer M9 images over M240 images it is because of different processing rather than because of inherent sensor characteristics?

Yes, although the Bayer CFA is made as a part of the sensor (similar on both CCD and CMOS), and they may well be different for the two sensors, requiring a bit different processing (which would also explain non-subjective differences).

Cheers,
Bart

[JV]

Yes, although the Bayer CFA is made as a part of the sensor (similar on both CCD and CMOS), and they may well be different for the two sensors, requiring a bit different processing (which would also explain non-subjective differences).

Cheers,
Bart

Interesting.  Thanks for explaining!

One of the characteristics often assigned to the CCD look is that files are more film-like.  A lot of CCD aficionados feel that CMOS files tend to look too smooth and too digital.

Would that be a consequence of the different processing?  Or is the above-mentioned lower signal to noise ratio that makes CCD files look less smooth?

[ErikKaffehr]

Personally, I am pretty sure that is just a myth. Quite true, CCDs generally have more readout noise, but at least decent exposures at base ISO are dominated by shot noise, the natural variation of photons.

Another question is, what is film like. Like Velvia or like Tri-X?

One explanation may be that film has sort of a gradual roll off in the highlights. With digital sensors only the linear portion of the transmission curve is used. Now, it seems that MFD-s are doing a lot of effort to keep exposure from saturation on the sensor. Things like overrating ISO, conservative on camera displays and histograms. When using the default "film curve" Capture One seems to overexpose the image something like 1-1.5 EV. That can push down exposure a bit, thus increasing shot noise somewhat.

Best regards
Erik

Interesting.  Thanks for explaining!

One of the characteristics often assigned to the CCD look is that files are more film-like.  A lot of CCD aficionados feel that CMOS files tend to look too smooth and too digital.

Would that be a consequence of the different processing?  Or is the above-mentioned lower signal to noise ratio that makes CCD files look less smooth?