Got my hands on the new Pentax 654Z today...

[Paul2660]

Torger

Back to your C1 & Raw Terapee side by sides. By default C1 consistently applies too much noise reduction even on base ISO files. You may try removing as much as 50 percent of the noise reduction. Also adding structure will help get the pastel look removed.

I often see this look on my Phase One conversions especially on distant details.  C1 allow for a local adjustment layer where you can add more structure to just those types of areas. It will make a big difference. I don't add any clarity just structure.

Then a final pass with Focus Magic and results look very good.

Paul

[BernardLanguillier]

Back to your C1 & Raw Terapee side by sides. By default C1 consistently applies too much noise reduction even on base ISO files. You may try removing as much as 50 percent of the noise reduction. Also adding structure will help get the pastel look removed.

Indeed, on my D800 files at ISO100 I always set all noise reduction to zero in C1 Pro.

Cheers,
Bernard

[Steve Hendrix]

Torger

Back to your C1 & Raw Terapee side by sides. By default C1 consistently applies too much noise reduction even on base ISO files. You may try removing as much as 50 percent of the noise reduction. Also adding structure will help get the pastel look removed.

I often see this look on my Phase One conversions especially on distant details.  C1 allow for a local adjustment layer where you can add more structure to just those types of areas. It will make a big difference. I don't add any clarity just structure.

Then a final pass with Focus Magic and results look very good.

Paul

Yes, and everyone should also keep in mind you can set the NR defaults in C1 to off with your camera. I have mine set that way as I prefer to always start with no luminance noise reduction at all, regardless of the ISO used. Specifically, I set Luminance NR to Zero, and leave the Color and Single Pixel settings at their standard default, as I rarely find their starting points to be a problem.


Steve Hendrix
Capture Integration

[Bart_van_der_Wolf]

Yes, and everyone should also keep in mind you can set the NR defaults in C1 to off with your camera.

Steve is correct. The camera defaults can be changed, and I have mine set to zero noise reduction. Capture One also allows to adjust the level of Detail it attempts to pull out of the noise, another trade-off.

Cheers,
Bart

[torger]

I re-ran my RT vs C1 test and turned off noise reduction -- virtually no difference in result, but then I looked at the default sharpening, and that's were the major difference is. The default USM of C1 has a 1.0 diameter compared to RT's 0.5. The larger diameter flattens the result making that look I find pastel-like at pixel peep. I guess C1's default is made for less sharp lenses or something.

Anyway, with the same type of sharpening the RT vs C1 result is very similar, I'd say that RT's Amaze is still a little bit ahead in terms of micro detail, but the difference is small so I'd say it's not relevant.

In other words, the difference in demosaicer result was not as large as I first thought.

[bjanes]

Hi Edmund,

I would not generalize that to all CMOS sensors having such circuits and settings. Most photographic sensor related ADCs output very linear quantized values. It's measurable with very simple means, and I've done it for a number of cameras, all had a linear response curve for most of the tonal range (if we eliminate noise).

Cheers,
Bart

That has been my experience with the Nikon D800e, which uses a Sony chip. Here are the results for the green channels which clip at the raw value of 15785. In the first graph shown below, the exposure increments are 0.3 EV and the upper three shots are clipped. A plot of the non-clipped channels shows nearly perfect linearity.

Bill

[Doug Peterson]

I re-ran my RT vs C1 test and turned off noise reduction -- virtually no difference in result, but then I looked at the default sharpening, and that's were the major difference is. The default USM of C1 has a 1.0 diameter compared to RT's 0.5. The larger diameter flattens the result making that look I find pastel-like at pixel peep. I guess C1's default is made for less sharp lenses or something.

Anyway, with the same type of sharpening the RT vs C1 result is very similar, I'd say that RT's Amaze is still a little bit ahead in terms of micro detail, but the difference is small so I'd say it's not relevant.

In other words, the difference in demosaicer result was not as large as I first thought.

Did you crank the details slider (Which modifies parameters of the demosaicer toward more aggressive detail extraction) to max? I think you'll find the comparison moot if/when you do that (and match the noise reduction and sharpening).

The default sharpening radius in C1 is intended for the average commercial photographer, not a pixel-peeping technically oriented one (no slight intended, just identifying two markets). A radius of 0.5 is only appropriate for a narrow range of lens/aperture/sensor/ISO/subject-matter/aesthetics. You can easily increase or decrease the radius and then save that as the default moving forward.

For use with a modern digital back at low ISO you may find you wish to modestly lower the default threshold of sharpening as well, which is kept at 1 by default to avoid sharpening noise (which is nearly absent in a well exposed IQ250 file at low ISO).

If detail is your absolute priority, you have an incredibly sharp lens with a sensor free of an AA filter, and you don't mind a bit of grain then:
- luminance and single-pixel noise reduction to zero
- detail slider to max
- sharpening threshold to 0
- sharpening radius to 0.5
- sharpening amount to taste
- add structure to taste

Many commercial photographers would find the result overly gritty and abrasive, but those who thrive off pixel detail should enjoy it quite a bit.

[written of course for more than just Torger, as he knows all this already]

[Bart_van_der_Wolf]

That has been my experience with the Nikon D800e, which uses a Sony chip. Here are the results for the green channels which clip at the raw value of 15785. In the first graph shown below, the exposure increments are 0.3 EV and the upper three shots are clipped. A plot of the non-clipped channels shows nearly perfect linearity.

Indeed, also almost perfectly linear. The slight onset of the slope in the most upper 1/3rd stop is usually due to the upper tail of the shot-noise starting to get clipped by saturation. Other noise sources, such as PRNU can be eliminated by plotting subtracted exposure pairs, but these shots already look amazingly (with an R^2 of 0.9999) linear to exposure.

Cheers,
Bart

[bjanes]

Indeed, also almost perfectly linear. The slight onset of the slope in the most upper 1/3rd stop is usually due to the upper tail of the shot-noise starting to get clipped by saturation. Other noise sources, such as PRNU can be eliminated by plotting subtracted exposure pairs, but these shots already look amazingly (with an R^2 of 0.9999) linear to exposure.

Cheers,
Bart

Yes, that upper 1/3 shot does show clipping of the shot noise, as shown by the standard deviation dropping from around 300 to around 30 and then to zero as the channel is totally clipped. I did subtract frames to eliminate PRNU and obtain an electron count, but I did not pursue the matter further as the results were already linear to a large degree. I only extracted the standard deviation of the subtracted images and not the mean. Would that mean add anything significant?

Bill

[torger]

[written of course for more than just Torger, as he knows all this already]

I don't use C1 much, just now and then, so these tips are useful to me too, thanks! Details slider did not make much of a difference, but yes I'd say with all things put together the difference in demosaicing performance is negligible. C1 does a great job just as RT.

As a landscape photographer using a 33 megapixel back I would guess that I have different needs of detail extraction than a portrait photographer using an 80 megapixel back, so the defaults are understandable.

What I try to achieve as a starting point is an image that looks as natural as possible at 100%, sharp but not over-sharpened or pastel-like or otherwise "digital"-looking, so the "structure" slider is not for me. Then depending on printer I do additional sharpening to match that.

Handled right, I think those 33 megapixels actually is enough in terms of detail for high end landscape photography, as reasonable depth of fields impose a certain limit on resolving power. The strongest reason for me to have more pixels is to battle aliasing rather than to get more resolving power.

(This thread must have set some sort of record in off-topicness  )

[ErikKaffehr]

Hi,

Although I also feel that we deviate to much from the OP, I think that this is interesting and a worthwhile issue to discuss. So I add a few observations of my own.

1) I have seen the phenomena you mention in C1 on my P45. It may be a combination of several effects, excessive NR and sharpening philosophy.

2) I have looked quite a lot into sharpness with different print sizes I have some objections about sharpening for actual pixels.

The reason is that we never see an image like pixel peeping on screen. What actually shows details best in real life is a print. But, sharpness in prints is dominated by lower frequencies like 10-20 lp/mm. So sharpening optimally at the pixel level may sharpen detail that has very little effect on perception of an actual print, unless the print is very large and viewed very close. So, C1 sharpening may be pretty good even if it may be excessive in actual pixel view.

For real good sharpness, I think the methods that Bart uses are probably best. No sharpening in the raw developer followed by very good sharpening with FocusMagic and some Topaz tools. The great benefit of FocusMagic is that it sharpens with very little halo.

I also found that sharpening optimally at actual pixels and applying some additional sharpening with USM, like radius 2 and amount 15% pushes low frequency detail close to 100%.

I would suggest that optimal sharpening is worth a separate thread.

Best regards
Erik

I don't use C1 much, just now and then, so these tips are useful to me too, thanks! Details slider did not make much of a difference, but yes I'd say with all things put together the difference in demosaicing performance is negligible. C1 does a great job just as RT.

As a landscape photographer using a 33 megapixel back I would guess that I have different needs of detail extraction than a portrait photographer using an 80 megapixel back, so the defaults are understandable.

What I try to achieve as a starting point is an image that looks as natural as possible at 100%, sharp but not over-sharpened or pastel-like or otherwise "digital"-looking, so the "structure" slider is not for me. Then depending on printer I do additional sharpening to match that.

Handled right, I think those 33 megapixels actually is enough in terms of detail for high end landscape photography, as reasonable depth of fields impose a certain limit on resolving power. The strongest reason for me to have more pixels is to battle aliasing rather than to get more resolving power.

(This thread must have set some sort of record in off-topicness  )

[torger]

The reason is that we never see an image like pixel peeping on screen.

I know this, that's why it's only a starting point. I believe that an image that looks natural at 100% on screen has a better enlargement potential than one that do not.

So far I make C-prints and for that process I find that print sharpening with some slight halo is best, as the printer kills the halo. But I find using 1.0 diameter is too large also for that, it does not make most out of the back's resolution. I do peep the prints too.

I haven't really worked with low frequency sharpening (other than the typical local contrast increases you can do also on screen), seems like an interesting subject. My guiding principle is natural look though, rather than something that appears as sharp as possible. Many guides I've read seems to be a bit "the more processing the better" rather than stopping when the image looks good and natural.

[Bart_van_der_Wolf]

Yes, that upper 1/3 shot does show clipping of the shot noise, as shown by the standard deviation dropping from around 300 to around 30 and then to zero as the channel is totally clipped. I did subtract frames to eliminate PRNU and obtain an electron count, but I did not pursue the matter further as the results were already linear to a large degree. I only extracted the standard deviation of the subtracted images and not the mean. Would that mean add anything significant?

Hi Bill,

Assuming that the exposure pairs were exposed to approx. the same level, after subtraction the mean would be close to zero and thus have little informational value. It's the random noise we're after. The means before subtraction would allow to verify shutter repeatability, and raw tonecurve linearity, and gain (by comparing the  number of input Photons with the resulting ADUs).

The subtraction is indeed essential if one wants to eliminate systematic noise (pattern noise, PRNU, hot/dead pixels, dust, etc.). After subtraction (and division by Sqrt(2) like you did) one is left with random dark-current+read-noise+shot noise, the latter of which dominates at high exposure levels. Assuming a Poisson noise distribution (if we wanted we could verify that by subtracting a Poisson noise distribution model, which should leave white noise as a residue), we can then reconstruct the average number of photons required to produce that level of noise, by squaring the noise Standard deviation.

Cheers,
Bart

[Bart_van_der_Wolf]

I know this, that's why it's only a starting point. I believe that an image that looks natural at 100% on screen has a better enlargement potential than one that do not.

Hi,

Given the current small number of actual 645Z users that can contribute, the thread is obviously drifting a bit OT, but such is life.

I'm known to be a huge supporter of proper Capture sharpening. Unfortunately most Raw converters offer little assistance to achieve that essential task accurately. Proper Capture sharpening attempts to only removes the optical/mechanical blur that is inherent to the the capture process. No halos should be created in that restoration process, only restoration of original signal sharpness. The blur is predominantly characterized by hardware (lens+IR/OLPF filter stack+sensel aperture) at this stage, with a small amount (only some 6.4%) of demosaicing compromises/trade-off added.

That Capture sharpened image data is the basis for further image contrast/acutance manipulation, global and local contrast. That part is known as Creative 'sharpening' and is largely influenced by the final viewing conditions/distance, and thus takes a bit of imagination and experience. Output sharpening then attempt to pre-compensate for later losses due to the output media.

I haven't really worked with low frequency sharpening (other than the typical local contrast increases you can do also on screen), seems like an interesting subject. My guiding principle is natural look though, rather than something that appears as sharp as possible. Many guides I've read seems to be a bit "the more processing the better" rather than stopping when the image looks good and natural.

I'd suggest to give the Topaz Labs Detail plugin a serious look. It allows to attenuate or boost several levels of detail, based on size relative to the total image dimensions, and offers options to modify luminance contrast based on complementary color. Maybe the best thing is that it allows to do so without creating halos, and while keeping colors looking natural (Intellicolor technology). This tool allows to stay in control and it offers lots of creative influence. It's one of the essential plugins in my toolchest (along side with FocusMagic for capture sharpening and Topaz Clarity for tonal contrast adjustment).

Cheers,
Bart

[LKaven]

Yes, that upper 1/3 shot does show clipping of the shot noise, as shown by the standard deviation dropping from around 300 to around 30 and then to zero as the channel is totally clipped. I did subtract frames to eliminate PRNU and obtain an electron count, but I did not pursue the matter further as the results were already linear to a large degree. I only extracted the standard deviation of the subtracted images and not the mean. Would that mean add anything significant?

Bill

The IQ-250, using the same sensor, does an automatic and mandatory dark frame subtraction on every exposure, no exceptions.  I'm wondering whether the Pentax (and the Hasselblad) also do the same thing?  I suspect that with a chip that has this much active circuitry and generates this much thermal noise, a dark frame subtraction just might be strongly recommended in the implementation notes.  Does your output suggest that this is occurring? 

I've often thought that my D800 ought to have an option to do a dark frame subtraction full-time, certainly at high gain settings.

[Doug Peterson]

The IQ-250, using the same sensor, does an automatic and mandatory dark frame subtraction on every exposure, no exceptions.  I'm wondering whether the Pentax (and the Hasselblad) also do the same thing?  I suspect that with a chip that has this much active circuitry and generates this much thermal noise, a dark frame subtraction just might be strongly recommended in the implementation notes.  Does your output suggest that this is occurring?  

This is not correct.

The IQ2 series does not have a user-option to modify the behavior of the dark frame. But it is not taken on every exposure; only when needed. In fact in many temperature/shutter-speed environments it is skipped for many frames at a time, including on multi-second exposures. e.g. If you do several 3 second exposures in a row in a temperature environment it's often only the first exposure which will get a dark frame after it.

[bjanes]

The IQ-250, using the same sensor, does an automatic and mandatory dark frame subtraction on every exposure, no exceptions.  I'm wondering whether the Pentax (and the Hasselblad) also do the same thing?  I suspect that with a chip that has this much active circuitry and generates this much thermal noise, a dark frame subtraction just might be strongly recommended in the implementation notes.  Does your output suggest that this is occurring? 

I've often thought that my D800 ought to have an option to do a dark frame subtraction full-time, certainly at high gain settings.

My tests were all at ISO100 with dark frame subtraction turned off. My longest exposure was for 5 minutes. Dark current was minimal and confined largely to a few hot pixels. Rawdigger stats for most of the image (excluding the masked pixels at the right of the frame) and the histogram are shown.

Bill

[LKaven]

This is not correct.

The IQ2 series does not have a user-option to modify the behavior of the dark frame. But it is not taken on every exposure; only when needed. In fact in many temperature/shutter-speed environments it is skipped for many frames at a time, including on multi-second exposures. e.g. If you do several 3 second exposures in a row in a temperature environment it's often only the first exposure which will get a dark frame after it.

Wait Doug, I did not say that the camera took a new dark frame on every exposure.  I said that it performed a dark frame subtraction on every exposure.  And I said that this was mandatory and cannot (and probably should not) be defeated.  My source, by the way, was you.  And as you confirmed here, I had my facts straight.

[Doug Peterson]

Wait Doug, I did not say that the camera took a new dark frame on every exposure.  I said that it performed a dark frame subtraction on every exposure.  And I said that this was mandatory and cannot (and probably should not) be defeated.  My source, by the way, was you.  And as you confirmed here, I had my facts straight.

Ha. You're completely right. I misread your post and skipped right over the word "subtraction". My appologies. After all, if you're sourcing from me who am I to disagree!