Will Canon Answer the D4s? [CR2]

[privatebydesign]

I use the 32bit technique a lot.

Shooting and creating photo real HDR

You can edit them in LR5 too, all the power and controls of ACR with 100% history.

 

[jrista]

I use the 32bit technique a lot.

Shooting and creating photo real HDR

You can edit them in LR5 too, all the power and controls of ACR with 100% history.

Thanks! That's actually really awesome, and extremely strait forward. I had no idea ACR would open a 32bit TIFF and give you 20 stops of editing latitude. Well, I'm going to have to play with that once spring and summer roll around, and I'm able to get back into landscapes. This is basically the kind of HDR editing I've always wanted, ironic that it was right at my fingertips...

 

[GMCPhotographics]

I do something a little different. I try to contain the whole contrast range in a single frame if possible.
If not, I'll shoot two or three frames to contain the range needed and then edit in LR, 16 bit files exported into CS5. These I lay over each other as layers and then I use a large soft eraser / brush and work on the layers until I get an image with no blown highlights or blocked up blacks....but still looks natural and simualr to how my eye remembers the scene.

 

[privatebydesign]

It isn't a panacea, and as I said in my reply a few days ago, I use it along with many other techniques. The key is to capture the information at the highest quality you can, then find the best way to present that information.

My top three methods for interiors with problematic exterior light levels are; 1, bring the light up in the room with strobes, this can look ok if the room is smaller and the architectural lighting is not a key element to the actual space. 2, layer mask in PS. 3, PS 32bit.

I often end up using a combination of all three but will bracket the hell out of the scene to guarantee I have all the information in front of me.

On one thread about HDR techniques many months ago I wrote about all the different kinds of blending, HDR, etc techniques and posted a same image comparison, but it is one of many hundreds of my posts that disappeared for no known reason. It was funny, the thread was basically about Photomatix and how horrible it was, yet my Photomatix HDR rendition, not the blend, was very popular and showed none of the comedy HDR look we are all so familiar with. Photomatix is an extremely powerful program and has recently been updated, skilful use of it can make remarkably subtle images now.

This again highlights another aspect of forum anxiety, I regularly shoot in situations where I am severely low iso DR limited, not just by my Canon's DR range even the mighty D800 wouldn't help much, because of that I have fully explored all the techniques available to mitigate that and am aware of how to work around it. Meanwhile other posters are adamant there is no DR "issue". However, I am not placing myself on a higher plane, I am a noise nazi, can't stand it, as far as I am concerned even a 7D at base is too noisy, but only because I haven't fully explored every noise reduction technique to satisfy myself, because it is not a priority for me. For the birders out there who have to shoot at >800iso they do prioritise that technique so don't understand my frustration at low levels of noise. Yes I have Topaz etc, but you can't know everything about every technique, and as far as end result, what is acceptable to one person might not be to another.

 

[jrista]

It isn't a panacea, and as I said in my reply a few days ago, I use it along with many other techniques. The key is to capture the information at the highest quality you can, then find the best way to present that information.

My top three methods for interiors with problematic exterior light levels are; 1, bring the light up in the room with strobes, this can look ok if the room is smaller and the architectural lighting is not a key element to the actual space. 2, layer mask in PS. 3, PS 32bit.

I often end up using a combination of all three but will bracket the hell out of the scene to guarantee I have all the information in front of me.

On one thread about HDR techniques many months ago I wrote about all the different kinds of blending, HDR, etc techniques and posted a same image comparison, but it is one of many hundreds of my posts that disappeared for no known reason. It was funny, the thread was basically about Photomatix and how horrible it was, yet my Photomatix HDR rendition, not the blend, was very popular and showed none of the comedy HDR look we are all so familiar with. Photomatix is an extremely powerful program and has recently been updated, skilful use of it can make remarkably subtle images now.

This again highlights another aspect of forum anxiety, I regularly shoot in situations where I am severely low iso DR limited, not just by my Canon's DR range even the mighty D800 wouldn't help much, because of that I have fully explored all the techniques available to mitigate that and am aware of how to work around it. Meanwhile other posters are adamant there is no DR "issue". However, I am not placing myself on a higher plane, I am a noise nazi, can't stand it, as far as I am concerned even a 7D at base is too noisy, but only because I haven't fully explored every noise reduction technique to satisfy myself, because it is not a priority for me. For the birders out there who have to shoot at >800iso they do prioritise that technique so don't understand my frustration at low levels of noise. Yes I have Topaz etc, but you can't know everything about every technique, and as far as end result, what is acceptable to one person might not be to another.

Well, there can most definitely be DR issues in many kinds of scenes. Photographing interiors is probably the best way to exacerbate dynamic range to the ultimate limits. If that's what you do, you should be screaming for Canon to reduce read noise to less than 1e-, and demand full 16-bit output with 20 stops of DR.

As for the 7D, by todays standards, there isn't any question it's a noisy little bugger. It's sweet spot is ISO 400-1600, it does rather poorly at ISO 100 and pretty badly at ISO 3200. It isn't TERRIBLY far behind modern APS-C cameras, however it really lags pretty far behind modern FF cameras. Between the 7D and 5D III or 6D, the difference in noise is so massive, it's almost painful. Part of that is due to the 7D having smaller pixels, part of it has to do with the strong AA filter (softer detail is more prone to being perceptually noisier, vs. sharper detail with is perceived as less noisy, even when they technically have the same amounts of noise)...however the 7D's pixels seem to have a particularly low full well capacity. At 4.3µm the FWC is 20187e-, where as the 70D with even smaller pixels has a FWC of 26726e-. The difference in Q.E. is a mere 4%, where as the difference in full well capacity is 32%! I am not sure what Canon did with the 70D sensor, but they have clearly made some fairly significant advancements with their sensor fabrication as far as efficiently utilizing the minimal amount of space they have. I would love to see ChipWorks break down the 70D sensor design...a die shrink to 180nm from 500nm could account for the necessary increase in surface area to support a higher FWC.

Even more impressive is the D7100, which as a FWC of 29236e- for pixels that are even smaller than the 70D. Again, the difference in Q.E. is relatively small, 11%, but the difference in charge capacity is huge, 45%! Canon's 18mp APS-C sensor simply wasn't really a thing of wonder...never really has been. It's surprising it's survived as long as it has in as many cameras as it has, given it's rather lackluster low-level specs. I am hoping the 70D sensor becomes the new standard for low end DSLRs and the EOS-M. I also sincerely hope the 7D II gets a sensor with a FWC in the 30,000e- range at 24mp...that would help immensely with the noise issues.

 

[DanielW]

It isn't a panacea, and as I said in my reply a few days ago, I use it along with many other techniques. The key is to capture the information at the highest quality you can, then find the best way to present that information.

My top three methods for interiors with problematic exterior light levels are; 1, bring the light up in the room with strobes, this can look ok if the room is smaller and the architectural lighting is not a key element to the actual space. 2, layer mask in PS. 3, PS 32bit.

I often end up using a combination of all three but will bracket the hell out of the scene to guarantee I have all the information in front of me.

On one thread about HDR techniques many months ago I wrote about all the different kinds of blending, HDR, etc techniques and posted a same image comparison, but it is one of many hundreds of my posts that disappeared for no known reason. It was funny, the thread was basically about Photomatix and how horrible it was, yet my Photomatix HDR rendition, not the blend, was very popular and showed none of the comedy HDR look we are all so familiar with. Photomatix is an extremely powerful program and has recently been updated, skilful use of it can make remarkably subtle images now.

This again highlights another aspect of forum anxiety, I regularly shoot in situations where I am severely low iso DR limited, not just by my Canon's DR range even the mighty D800 wouldn't help much, because of that I have fully explored all the techniques available to mitigate that and am aware of how to work around it. Meanwhile other posters are adamant there is no DR "issue". However, I am not placing myself on a higher plane, I am a noise nazi, can't stand it, as far as I am concerned even a 7D at base is too noisy, but only because I haven't fully explored every noise reduction technique to satisfy myself, because it is not a priority for me. For the birders out there who have to shoot at >800iso they do prioritise that technique so don't understand my frustration at low levels of noise. Yes I have Topaz etc, but you can't know everything about every technique, and as far as end result, what is acceptable to one person might not be to another.

Well, there can most definitely be DR issues in many kinds of scenes. Photographing interiors is probably the best way to exacerbate dynamic range to the ultimate limits. If that's what you do, you should be screaming for Canon to reduce read noise to less than 1e-, and demand full 16-bit output with 20 stops of DR.

As for the 7D, by todays standards, there isn't any question it's a noisy little bugger. It's sweet spot is ISO 400-1600, it does rather poorly at ISO 100 and pretty badly at ISO 3200. It isn't TERRIBLY far behind modern APS-C cameras, however it really lags pretty far behind modern FF cameras. Between the 7D and 5D III or 6D, the difference in noise is so massive, it's almost painful. Part of that is due to the 7D having smaller pixels, part of it has to do with the strong AA filter (softer detail is more prone to being perceptually noisier, vs. sharper detail with is perceived as less noisy, even when they technically have the same amounts of noise)...however the 7D's pixels seem to have a particularly low full well capacity. At 4.3µm the FWC is 20187e-, where as the 70D with even smaller pixels has a FWC of 26726e-. The difference in Q.E. is a mere 4%, where as the difference in full well capacity is 32%! I am not sure what Canon did with the 70D sensor, but they have clearly made some fairly significant advancements with their sensor fabrication as far as efficiently utilizing the minimal amount of space they have. I would love to see ChipWorks break down the 70D sensor design...a die shrink to 180nm from 500nm could account for the necessary increase in surface area to support a higher FWC.

Even more impressive is the D7100, which as a FWC of 29236e- for pixels that are even smaller than the 70D. Again, the difference in Q.E. is relatively small, 11%, but the difference in charge capacity is huge, 45%! Canon's 18mp APS-C sensor simply wasn't really a thing of wonder...never really has been. It's surprising it's survived as long as it has in as many cameras as it has, given it's rather lackluster low-level specs. I am hoping the 70D sensor becomes the new standard for low end DSLRs and the EOS-M. I also sincerely hope the 7D II gets a sensor with a FWC in the 30,000e- range at 24mp...that would help immensely with the noise issues.

Another brilliant post, jrista!
Learning a lot from you, neuro, Mt Spokane and others. Thank you all.
Daniel

 

[Orangutan]

It is impossible to have more than 14 stops of actual real-world DR with DSLRs because we only have 14-bit AD converters.

At the risk of looking like an idiot, I confess I don't quite follow this. I've never examined de-mosaic algorithms, but I haven't assumed that they are limited to weighted averages of adjacent photosites. I've always assumed that a de-mosaic algorithm could choose to include some "addition." In other words, that the resultant pixel RGB values could be scaled up to accommodate adjacent photosites that are (nearly) maxed out or nearly-zero. By analogy, it's the difference between rolling two dice and averaging, vs. rolling two dice and adding.

Certainly an individual photosite can't have more DR than permitted by its physical characteristics, but I don't see why the resultant post-demosaic scale can't, as it were, "go up to 11."

 

[jrista]

It is impossible to have more than 14 stops of actual real-world DR with DSLRs because we only have 14-bit AD converters.

At the risk of looking like an idiot, I confess I don't quite follow this. I've never examined de-mosaic algorithms, but I haven't assumed that they are limited to weighted averages of adjacent photosites. I've always assumed that a de-mosaic algorithm could choose to include some "addition." In other words, that the resultant pixel RGB values could be scaled up to accommodate adjacent photosites that are (nearly) maxed out or nearly-zero. By analogy, it's the difference between rolling two dice and averaging, vs. rolling two dice and adding.

Certainly an individual photosite can't have more DR than permitted by its physical characteristics, but I don't see why the resultant post-demosaic scale can't, as it were, "go up to 11."

That would be post-demosaic, though. I mean, we can downsample images and gain DR as well...but again, post-demosaic. Our editing latitude in a tool like Lightroom comes from editing the RAW image. The RAW is effectively a digital signal, and we push that signal around with the exposure, highlight and shadow sliders, the tone curve, etc. Once you "rasterize" that digital signal, your editing latitude disappears. If you've ever tried to push a 16-bit TIFF around the same way you push a RAW around, you would understand how much you lose by converting to an RGB image...you lose a LOT. So, technically speaking, you could gain a stop or more of dynamic range simply by downsampling...however in order to downsample, you have to demosaic the RAW...and you lose your editing latitude. You have less noise, but you don't have quite the editability you once did.

In the statement of mine you quited, I was speaking about the hardware aspects of a camera. The RAW, as it comes strait out of a DSLR, is limited in terms of dynamic range, by the bit depth of the ADC. If you have a 14-bit ADC, you can't have more than 14 stops of DR...and to actually achieve 14 stops of DR, you would require a zero noise floor. Not even Sony Exmor has a zero noise floor, which is why DXO's "Screen DR" measure says 13.2 stops, rather than 14.4 stops, of dynamic range. DXO Screen DR is effectively a direct measure of the hardware capabilities...Print DR is a measure of a post-processed image (and, to be quite frank, we don't know exactly what kind of processing is used to produce the 12mp normalized "print" image they measure Print DR from...so it is, IMO, a sketchy measure.)

For all intents and purposes, until we have 16-bit ADC, the theoretical maximum hardware DR of a camera is 14 stops, and due to noise, the actual realizable dynamic range is going to be less than 14 stops.

 

[Orangutan]

That would be post-demosaic
(snip)
If you've ever tried to push a 16-bit TIFF around the same way you push a RAW around,

But isn't everything on a Bayer sensor post-demosaic? You can use your slider settings as parameters to your demosaic algorithm, but you still have to demosaic the raw file to have an "image" rather than "data." E.g. a very intense, pure-green light will show as (near) zero on a red-filtered photosite.

downsample

I get the downsample thing, and I don't dispute that.

how much you lose by converting to an RGB image...you lose a LOT.

Seems to me this could have more to do with round-off error than raw vs. TIFF. I'm not, by any measure, an expert in Photoshop, but maybe I'll try "pushing around" a TIFF file using adjustment layers vs. old-style hard edits.

In the statement of mine you quited, I was speaking about the hardware aspects of a camera. The RAW, as it comes strait out of a DSLR, is limited in terms of dynamic range, by the bit depth of the ADC.

I guess this is another item I don't quite get: I'd assume "hardware" DR would be limited by the fwc and read noise, rather than the ADC. You could always have an ADC whose digital output does not fall on EV boundaries, being either coarser or finer. E.g., the 20D has fwc of about 51,400e (per Clarkvision) but has 12-bit ADC. The 7D has fwc of about 24,800e (per Clarkvision) but has a 14-bit ADC. If I understand correctly, this should give the 7D finer gradation, but not necessarily more hardware DR. (all dependent on noise, of course)

 

[privatebydesign]

Re the linked video above, you don't need CS6, CS5 will do it too but you can't open the file it creates in PS5, you can in LR5.

So, CS5 + LR5 and 32 bit works as you can edit the file in LR, 32 bit compatibility was added to LR5. LR4 and CS6, same thing, you just open the file in CS6 but can "store" it in LR4 and open the original files to remap in LR4.

CS6 and LR5 and you are golden, you can adjust the file anywhere. When you have a 32 bit file in LR5 the exposure slider changes to +/-10, the other sliders stay the same values.

 

[jrista]

That would be post-demosaic
(snip)
If you've ever tried to push a 16-bit TIFF around the same way you push a RAW around,

But isn't everything on a Bayer sensor post-demosaic? You can use your slider settings as parameters to your demosaic algorithm, but you still have to demosaic the raw file to have an "image" rather than "data." E.g. a very intense, pure-green light will show as (near) zero on a red-filtered photosite.

No. When you edit a raw, most of the exposure settings, white balance, and a number of other things are applied to the RAW data. Demosaicing converts the results of that initial processing on the RAW into an image that can be displayed on screen. Other settings, such as sharpening, occur after demosaicing (which is partly why sharpening in a RAW editor can have such a great impact on increasing noise.) Some raw editors apply wavelet denoise algorithms on the RAW, and a couple tools like Topaz DeNoise apply some pre and some post demosaic denoise if you apply it to the raw.

Most RAW settings are applied pre- or during- demosaic, and some are applied after. Demosaicing, during RAW editing, is near the end of the pipeline that repeatedly renders, realtime, a RAW image on screen. If you export to a TIFF, this same pipeline is used to render to a file rather than to the screen. Any edits on the TIFF, from that point on, are obviously on RGB data.

downsample

I get the downsample thing, and I don't dispute that.

how much you lose by converting to an RGB image...you lose a LOT.

Seems to me this could have more to do with round-off error than raw vs. TIFF. I'm not, by any measure, an expert in Photoshop, but maybe I'll try "pushing around" a TIFF file using adjustment layers vs. old-style hard edits.

It doesn't really matter if you use old style edits or adjustment layers. Once demosaiced, you lose your editing latitude. It's like using sRAW or mRAW in Canon cameras...you can push exposure around a bit, but you lose a lot of editing latitude. Where before you might be able to shift exposure up or down by five stops without encountering attenuation limitations or artifacts, with sRAW, you can only push exposure around a couple stops before you run into problems. TIFF is the same deal...you can push levels and curves around a bit, but attenuate too much, and you run into problems.

In the statement of mine you quited, I was speaking about the hardware aspects of a camera. The RAW, as it comes strait out of a DSLR, is limited in terms of dynamic range, by the bit depth of the ADC.

I guess this is another item I don't quite get: I'd assume "hardware" DR would be limited by the fwc and read noise, rather than the ADC. You could always have an ADC whose digital output does not fall on EV boundaries, being either coarser or finer. E.g., the 20D has fwc of about 51,400e (per Clarkvision) but has 12-bit ADC. The 7D has fwc of about 24,800e (per Clarkvision) but has a 14-bit ADC. If I understand correctly, this should give the 7D finer gradation, but not necessarily more hardware DR. (all dependent on noise, of course)

Well, depends on what you consider hardware. If you are thinking purely sensor, then yes, DR is limited by FWC and read noise. According to some of Roger Clark's older work with Canon gear, their sensors themselves have actually been capable of around 15 stops of dynamic range for a while. The intrinsic read noise component from dark current in a Canon sensor is pretty low, only a couple electrons. Canon CDS technology, built into their sensors, is very good. However, the more significant component of read noise in Canon cameras comes from downstream sources. One of those sources is their high frequency off-die ADC units themselves, which probably consume about a stop or so worth of DR alone. Canon also has a downstream amplifier, used in a variety of situations, which also has the potential to add read noise. Shipping an analog signal along a bus is also another opportunity for noise to be introduced. Total read noise would be an amalgamation of noise from all these sources, not just the sensor itself.

When I refer to hardware DR, I am referring to this entire processing pipeline in the camera. I'd say Sensor DR if I meant just the sensor, but hardware DR is what you get out of the sensor + bus + downstream junk + ADC + DSP. Since the ADC is 14 bits, and the output is digital rather than analog, your implicitly limited to 14 stops, regardless of what the sensor itself might actually be capable of. Your quantizing the analog sensor information, with a cap on the maximum quantization value. If Canon can solve their downstream noise problems, if their intrinsic dark current noise from the sensor really is as low as 1.5-2e- after CDS, then I think Canon could actually benefit from 16-bit ADC. We might be able to get as much as 15.6 stops of DR out of Canon's current sensor tech.

 

[Orangutan]

RAW

Thanks for the extended response. I'm not quite convinced, but I'll chew on it a while.

 

[SiliconVoid]

I think the bigger question is - Does Canon Even Need to Answer the D4s...
(btw: if naming convention holds up the D4s will not be any significant increase in mp, if any at all, so it is not going to be a 'pro-body' D800 - just improvements Nikon feels worthy of a different model name)

If we look at Canon's current lineup of top tier bodies (1Dx, 5DmkIII, 6D, 70D, T5i) they all outsell their competitors offerings. While Canon surely wants to advance the technology as much as the next brand, it comes down to dollar and cents in the end.. and it is there that Canon continues to hold the lead. So other than bragging rights to some ambiguous scoring service there is actually little reason for Canon to worry about a specific brand or model they already outsell...

Not everyone will agree with this but as for the other argument - DR! - well it is not as significant or as necessary as many believe. By that I mean while a generous range can provide flexibility and creativity in certain situations it is of limited use. A properly or creatively exposed shot can relieve the need for 14 stop post processing as there would be no 'need' for it to begin with. To give an example: If Canon/Nikon/Sony/etc were to develop a sensor that captured every scene with well lit shadows, exaggerated colors, exaggerated contrast, a complete dream like scene with 50-stops of DR, and no more leeway for processing because the sensor has already captured and reproduced everything there is to be seen - people would still complain about the lack of stops they have in post as a must have, must design etc.

Isn't photography the art/science/creativity of capturing light and shadow?? If the scene being captured has shadows you cannot see into with your eye then there is no 'need' to remove the shadows in development/print, actually doing so tends to ruin the feel of the scene most of the time. It is true that there is an interest/intrigue in an image that looks the way you see the world in a dream - where everything is lit by some magical indirect lighting coming from every direction - but as it is not the way we see the world in real life it will always be a method of processing, a fad, an interest that comes and goes.. It is not something a camera manufacturer need redesign their products around. It is simply not something 'needed' to the point that people fret over 1,2,3 stops difference between this model or that brand.

We would all be better served by a sensor that produces noise free images through the ISO range, even if it did not capture any more than 11 stops of DR, and I mean without downsampling, without film-like-grain, and without 3x the post processing.. I personally would not care if Canon ever squeaked out another stop of DR as long as they work towards ISO 100 performance at >6400 ISO...

 

[mkabi]

I'm sure you guys still remember this :

To be honest, I'm one of those thought Canon going to have something in vintage as well. Last time I check, Df didn't do welllllllll

I think it didn't do well because of the borked controls, not because it was "vintage". I'd welcome a Canon vintage body design, so long as it did not include the hideous stacked dial controls and...well, basically kept the phenomenal electronic controls and button placement that is now standard on Canon pro bodies, just in a nostalgic retro body design. And, yes, with out any video features...at all... ;o)

Canon didn't exactly create it, but then again by this guys standards, Nikon is about 3 years too late:
http://www.engadget.com/2010/04/29/canon-ae-1-program-slr-gets-a-digital-retrofit/

 

[jrista]

I think the bigger question is - Does Canon Even Need to Answer the D4s...
(btw: if naming convention holds up the D4s will not be any significant increase in mp, if any at all, so it is not going to be a 'pro-body' D800 - just improvements Nikon feels worthy of a different model name)

If we look at Canon's current lineup of top tier bodies (1Dx, 5DmkIII, 6D, 70D, T5i) they all outsell their competitors offerings. While Canon surely wants to advance the technology as much as the next brand, it comes down to dollar and cents in the end.. and it is there that Canon continues to hold the lead. So other than bragging rights to some ambiguous scoring service there is actually little reason for Canon to worry about a specific brand or model they already outsell...

Not everyone will agree with this but as for the other argument - DR! - well it is not as significant or as necessary as many believe. By that I mean while a generous range can provide flexibility and creativity in certain situations it is of limited use. A properly or creatively exposed shot can relieve the need for 14 stop post processing as there would be no 'need' for it to begin with. To give an example: If Canon/Nikon/Sony/etc were to develop a sensor that captured every scene with well lit shadows, exaggerated colors, exaggerated contrast, a complete dream like scene with 50-stops of DR, and no more leeway for processing because the sensor has already captured and reproduced everything there is to be seen - people would still complain about the lack of stops they have in post as a must have, must design etc.

Isn't photography the art/science/creativity of capturing light and shadow?? If the scene being captured has shadows you cannot see into with your eye then there is no 'need' to remove the shadows in development/print, actually doing so tends to ruin the feel of the scene most of the time. It is true that there is an interest/intrigue in an image that looks the way you see the world in a dream - where everything is lit by some magical indirect lighting coming from every direction - but as it is not the way we see the world in real life it will always be a method of processing, a fad, an interest that comes and goes.. It is not something a camera manufacturer need redesign their products around. It is simply not something 'needed' to the point that people fret over 1,2,3 stops difference between this model or that brand.

We would all be better served by a sensor that produces noise free images through the ISO range, even if it did not capture any more than 11 stops of DR, and I mean without downsampling, without film-like-grain, and without 3x the post processing.. I personally would not care if Canon ever squeaked out another stop of DR as long as they work towards ISO 100 performance at >6400 ISO...

You clearly don't understand the primary source of noise. It is impossible to have ISO 100 performance at ISO 6400, while still having comparable sensor resolution to sensors of today. "Noise" is a general term that refers to ALL noise in an image. NOT all noise in an image is from the camera's electronics. Noise caused by camera electronics is called read noise, however read noise only affects the deep shadows, and it is generally only present to a relatively significant degree at lower ISO settings. You are also missing the fact that dynamic range is relative to noise. Eliminate noise, and you effectively have infinite dynamic range (or, in the case of a digitized result, you gain the maximum dynamic range up to your bit depth...whatever that may be...14bits/14stops, 16bits/16stops, 1024bits/1024stops.)

The primary source of noise, by a very significant margin, is photon shot noise. This noise is present in the analog image signal itself, and has absolutely NOTHING to do with the camera. The amount of photon shot noise is approximated by SQRT(Signal), so as signal strength drops (which is what happens when you crank up ISO), the ratio of noise to signal increases. Canon cannot fix that. Canon, as well as every other camera and sensor manufacturer on the planet, have absolutely no control over that. You will never have ISO 100 performance at very high ISO settings like ISO 6400. For that matter, even ISO 100 has noise...its less, but noise is always present in every signal, regardless of what the ISO setting is.

Even if Canon reduced megapixel count to 1mp, and greatly increased pixel size, you are STILL not going to have ISO 100 like performance...it'll be a lot better, but it will still be noisy relative to the image size, because the total signal strength at ISO 6400 is still the same, it's just spread out across fewer pixels with larger capacities. Reducing megapixel count is largely no different than downsampling. You gather more photons per pixel...you lose detail (in the case of a 1mp sensor, a LOT of detail), but you have less apparent noise. ISO 100 noise levels drop right along with ISO 6400 noise levels, so even though ISO 6400 is better, ISO 100 is that much better, too! If you take this concept to its ultimate conclusion, you eventually arrive at a one-pixel sensor of infinite size...that would be the only way to actually eliminate noise at all ISO settings...but, it's entirely impractical and implausible.

You MUST make a trade off. More megapixels, more per-pixel noise, fewer megapixels, less per-pixel noise. Doesn't matter if you reduce sensor pixel count, or downsample in post, either way, it's the same tradeoff. Assuming we stick with current technology, we could DOUBLE quantum efficiency for all sensors that have less than 50% (which isn't that many these days, most sensors are 49-51% Q.E. at least now). If we double quantum efficiency, and leave pixel size the same, that only means that ISO 6400 is now as good as ISO 3200. At the same time, ISO 100 got a stop better as well! It is now as good as a native ISO 50 would have been on a sensor with a Q.E. of 50%. Once were at 100% Q.E. (also technically infeasible...at best we could get somewhere around 90% or so with extreme cooling to -80°C), that's it...we cannot improve quantum efficiency any more. ISO 6400, for that sensor resolution, is the best it's going to be without taking some radical departure from standard sensor designs.

Some patents exist, like color splitting filters, as an alternative to color filter arrays. This might get you another half stop or so. So, ISO 6400 might look as good as ISO 2500. Maybe we employ some kind of layered photodiode...its been done in Foveon-type sensors, however its tricky and the impact on noise is minimal in practice. We might gain another half stop...so ISO 6400 now looks like ISO 1600. Back-illuminated sensor design might get us a small fraction of a stop for pixel sizes as big as they are in DSLRs...it wouldn't be worth the added fabrication costs. That about exhausts the extreme measures we could take to improve ISO 6400. Our sensors will now probably cost a good three to four times as much as they did if we employ all of these techniques...all for two stops of ISO improvement. Were still a long, long ways from ISO 6400 looking anything remotely as good as ISO 100, however...and there is still that nagging little fact that every time we improve ISO 6400, we also improve ISO 100. We never actually achieve the goal of normalizing noise at all ISO settings, because anything we do to make ISO 6400 better makes all the other ISO settings better as well.

It doesn't matter what you do, there is no normalizing the noise levels of different ISO settings. There will always be noise, at all ISO settings, and noise will increase as the square root of the signal as ISO is increased, because that's simply how the physics works. It is impossible to have the same levels of noise at all ISO settings. It's a matter of physics, not technology. We can't break the laws of physics. And they are already bent pretty far with current sensor technology...it is nothing short of amazing that we get the kind of IQ we currently do out of small form factor sensors with 1100nm pixels...that is as small as a wavelength of deep infrared light!!

Now, contrary to the issues above with eliminating all noise at all ISO settings, camera manufacturers DO have control over how much noise their electronics generate. They don't have total control, some things are still beyond their control...for example, we cannot completely eliminate dark current noise, but we can reduce it with CDS (Correlated Double Sampling), and we can greatly reduce it even more by cooling sensor circuitry to temperatures well below zero (-80°C is the sweet spot for power vs. dark current reduction). We can reorganize circuitry, move high frequency components into isolated areas on the die, increase parallelism and reduce operating frequency, and probably a whole host of other things that are currently being discovered or have yet to be discovered that give us control over read noise.

By controlling read noise, we reduce the thing that is actually eating away at dynamic range at lower ISO settings. Canon sensors are not limited to 11 stops of DR. Actually, according to some older studies done by I believe Roger Clark of Clarkvision, when we ignore downstream sources of read noise, Canon's current sensors are likely capable of over 15 stops of dynamic range in analog space. That dynamic range is REDUCED by read noise, which includes noise from dark current as well as noise from high frequency components downstream of the sensor. Canon technically has a lot of options when it comes to reducing this source of noise...hence the reason low ISO dynamic range is a highly contentious point with Canon users. Many manufacturers in the CIS industry have started moving past the 11-12 stop "barrier" that used to be the limit throughout the first half or so of the last decade. Several manufacturers are achieving more than 12 stops of DR at ISO 100, and one has achieved over 13 stops of DR at ISO 100. All of them have achieved that by reducing read noise.

So sorry...but Canon cannot eliminate all noise. Simply not possible. Canon CAN reduce the noise their camera electronics are introducing into the low end of the image signal, however that will do little to affect ISO 6400 performance. Even assuming we employ all the best known options for improving literal light sensitivity on the sensor, outside of GREATLY reducing pixel count (by a factor of two or more, which is really just the same as downsampling), we MIGHT get another two stops of noise performance before we hit an impenetrable brick wall. That still leaves us at least four stops away from having ISO 100 performance at ISO 6400. Assuming the trend towards higher megapixel counts continues, that will only continue to diminish performance at ISO 6400, meaning any technological improvements to improve light sensitivity will only restore ISO 6400 noise performance to the level of sensors with fewer megapixels.

 

[AvTvM]

... Assuming the trend towards higher megapixel counts continues, that will only continue to diminish performance at ISO 6400, meaning any technological improvements to improve light sensitivity will only restore ISO 6400 noise performance to the level of sensors with fewer megapixels.

while I really like to read all your insights I would still like to question your last sentence here a bit:

Has not the D800 sensor (+ electronics) proven, that it is possible to improve on both targets - resolution AND image quality [at low and high ISO] ... at the same time? The D800 has dramatically more resolution than the preceding D700 and still better (less visible noise and more DR, color fidelity) at ISO 6400.

While there obviously is some tradeoff between resolution, low-ISO IQ (DR, noise/banding) and hi-ISO IQ it seem that with innovative approaches all 3 corners of the triangle can be pushed forward. SOny (+ Nikon) have been able to do this and get products to market that deliver the goods to customers, while Canon has been lagging for some time by now. Canon has managed to only push 1 corner of this triangle: hi-ISO image-quality - but not resolution and low-ISO IQ. And even in the Hi-ISO IQ, Canon's sensors are definitely not really superior to Nikon/Sony sensors: 1DX sensor is really only marginally ahead of the D800 from ISO 3200 upwards and probably not at ahead of the D4.

As far as I am concered, that is the main source of "relative discontent" with Canon compared to Nikon (/Sony) when discussing sensor performance.

My feeling is, that Canon up to now really is UNABLE to match Nikon/Sony sensor performance, rather than UNWILLING to do so - or intentionally "holding back", saving improvements for future products.

 

[jrista]

... Assuming the trend towards higher megapixel counts continues, that will only continue to diminish performance at ISO 6400, meaning any technological improvements to improve light sensitivity will only restore ISO 6400 noise performance to the level of sensors with fewer megapixels.

while I really like to read all your insights I would still like to question your last sentence here a bit:

Has not the D800 sensor (+ electronics) proven, that it is possible to improve on both targets - resolution AND image quality [at low and high ISO] ... at the same time? The D800 has dramatically more resolution than the preceding D700 and still better (less visible noise and more DR, color fidelity) at ISO 6400.

While there obviously is some tradeoff between resolution, low-ISO IQ (DR, noise/banding) and hi-ISO IQ it seem that with innovative approaches all 3 corners of the triangle can be pushed forward. SOny (+ Nikon) have been able to do this and get products to market that deliver the goods to customers, while Canon has been lagging for some time by now. Canon has managed to only push 1 corner of this triangle: hi-ISO image-quality - but not resolution and low-ISO IQ. And even in the Hi-ISO IQ, Canon's sensors are definitely not really superior to Nikon/Sony sensors: 1DX sensor is really only marginally ahead of the D800 from ISO 3200 upwards and probably not at ahead of the D4.

As far as I am concered, that is the main source of "relative discontent" with Canon compared to Nikon (/Sony) when discussing sensor performance.

My feeling is, that Canon up to now really is UNABLE to match Nikon/Sony sensor performance, rather than UNWILLING to do so - or intentionally "holding back", saving improvements for future products.

The D800 is no better at high ISO than any other camera. My post was specifically addressing the notion that you could flatten the noise curve, or even eliminate noise. You can't. It is IMPOSSIBLE to simultaneously have ISO 100 and ISO 6400 perform the same. Cannot happen. If you reduce the quantity of light, the ratio of noise to signal will increase, simple as that...and that has nothing to do with the hardware, that is an attribute of gathering light.

The D800 did exactly what I said Canon could do: Reduce read noise. That's it. Sony Exmore has 3e- read noise at all ISO settings, which is why its LOW ISO settings are better. Canon has a non-linear read noise curve. In the case of the 5D III, for example, it is 33e- at ISO 100, 18.2e- at ISO 200, 10.6e- at ISO 400, and 6e- at ISO 800. After that it drops to 3e- and below. Canon needs to eliminate their read noise, and they will gain Low ISO DR as a result. But Canon can't really do anything to make ISO 6400 have the same amount of total noise as ISO 100, and they have very limited options to reduce high ISO noise in general (which is done by increasing signal strength, which allows for a reduction in gain, which is directly what impacts photon shot noise...but we are talking fractional changes here, nothing significant or particularly substantial.)

All cameras currently on the market that have pixels that fall into certain size ranges are going to have similar high ISO performance. There isn't even the option of making them significantly different, because it is a matter of physical limitations, not technological. Canon has a lead when it comes to high ISO performance over SoNikon right now, however it is marginal. The primary reason Canon has the lead is generally larger pixels, and better CDS (Canon's high ISO read noise can be less than 2e-, where as with SoNikon sensors it is usually closer to 3e-.) The D800 has visibly worse high ISO performance than any Canon sensor (especially the 6D, and except the old 18mp APS-C) due to its use of smaller pixels. And that is entirely expected by the theory...high ISO is limited by physics, and how that limit affects noise is ultimately guided by FWC...of which the D800 has 33% less of compared to the 5D III (and 41% less than the 6D).

 

[CarlTN]

Jrista, without quoting that long post...do you honestly think it's possible for a DSLR to have its sensor cooled to -80 C? How would that be done in a DSLR?

 

[jrista]

Jrista, without quoting that long post...do you honestly think it's possible for a DSLR to have its sensor cooled to -80 C? How would that be done in a DSLR?

Active cooling is actually done quite frequently today with high end astrophotography cameras. They use dual stage peltier (TE/TEC, or ThermoElectric) cooling. Peltiers are very thin electronic heat pumps, being simply an array of N and P type silicon sandwiched between two ceramic plates (one "cold" plate and one "hot" plate, heat is pumped from the cold side to the hot side). It would be easy to fit a peltier into existing DSLR bodies without anyone being the wiser (with the exception of increased heat output, as the peltier generates it's own heat along with drawing out heat from whatever is attached to it's cold side.) Now, with most astrophoto cameras, the delta-T they aim for is around 50°C. On an average night, the temperatures drop to somewhere around -10° to -20°, however the really high end ones can cool much more effectively with delta-T over 60°. On a cold night, CCD temp with a really good astrocam can get to below -75°. Most cooled astrocams also employ low noise fans to actively cool a heat sink or heat pipe attached to the hot side of the upper peltier and actively exhaust heat...something similar could be done with a DSLR.

Professional scientific grade CCD cameras used in professional astrophotography, microscopy, etc. use much more significant measures to cool. Professional scientific grade CCD cameras are usually cooled to at least -80°C. In some cases, temperatures are pushed below -125°C, and I've even heard of some scientific grade equipment operating in superconducting conditions at nearly absolute zero (however once you move past -80°C, the cost of maintaining temperature becomes excessively prohibitive.)

In the case of a DSLR, at some point I see some kind of peltier based cooling becoming necessary. At some point, we are going to exhaust the material options, when we've employed things like black silicon, color splitting in favor of color filtration arrays, and maybe even some kind of layered photodiode approach to increase maximum charge capacity per pixel. To continue improving (and at that point, low ISO will be about as good as it can get, so all the improvements will have to occur at the high ISO end), without reducing megapixel count, we will need to reduce dark current noise in the electronics themselves. The most effective way to reduce dark current once CDS is employed is to cool the sensor.

Even for a relatively cheap $2000 astrocam with dual-stage peltier cooling, average dark current drops from around 5e- to 0.02e-, and better ones can be had for $4000 to $10,000 where dark current drops to as little as 0.01e- to 0.008e-. At 0.01e-, you release one electron worth of noise for every 100 electrons released by photons. Today, average read noise at high ISO is around 3e- or so, so cooling could gain us a fair amount of real-world high ISO sensitivity. Even at medium ISO settings, where dark current can still be as high as 5-10e-, could benefit from cooling. Extreme cooling could even be an option to reduce ISO 100 noise as well, albeit at a power cost.

On the notion of power consumption, that would certainly be a hurdle to overcome. Power cells would have to be far more efficient, and certainly hold more capacity, than even the most capable camera batteries of today. I suspect some kind of fuel cell technology would need to be employed to make extreme peltier cooling a reality for high ISO shooters. Fuel cell tech has come a long way recently, and I suspect at some point camera manufacturers will probably switch to them anyway. Thermoelectric cooling could be a user-selectable option as well, and the peltier could be activated automatically on demand if it is enabled so that it does not constantly draw power.

 

[CarlTN]

Jrista, without quoting that long post...do you honestly think it's possible for a DSLR to have its sensor cooled to -80 C? How would that be done in a DSLR?

Active cooling is actually done quite frequently today with high end astrophotography cameras. They use dual stage peltier (TE/TEC, or ThermoElectric) cooling. Peltiers are very thin electronic heat pumps, being simply an array of N and P type silicon sandwiched between two ceramic plates (one "cold" plate and one "hot" plate, heat is pumped from the cold side to the hot side). It would be easy to fit a peltier into existing DSLR bodies without anyone being the wiser (with the exception of increased heat output, as the peltier generates it's own heat along with drawing out heat from whatever is attached to it's cold side.) Now, with most astrophoto cameras, the delta-T they aim for is around 50°C. On an average night, the temperatures drop to somewhere around -10° to -20°, however the really high end ones can cool much more effectively with delta-T over 60°. On a cold night, CCD temp with a really good astrocam can get to below -75°. Most cooled astrocams also employ low noise fans to actively cool a heat sink or heat pipe attached to the hot side of the upper peltier and actively exhaust heat...something similar could be done with a DSLR.

Professional scientific grade CCD cameras used in professional astrophotography, microscopy, etc. use much more significant measures to cool. Professional scientific grade CCD cameras are usually cooled to at least -80°C. In some cases, temperatures are pushed below -125°C, and I've even heard of some scientific grade equipment operating in superconducting conditions at nearly absolute zero (however once you move past -80°C, the cost of maintaining temperature becomes excessively prohibitive.)

In the case of a DSLR, at some point I see some kind of peltier based cooling becoming necessary. At some point, we are going to exhaust the material options, when we've employed things like black silicon, color splitting in favor of color filtration arrays, and maybe even some kind of layered photodiode approach to increase maximum charge capacity per pixel. To continue improving (and at that point, low ISO will be about as good as it can get, so all the improvements will have to occur at the high ISO end), without reducing megapixel count, we will need to reduce dark current noise in the electronics themselves. The most effective way to reduce dark current once CDS is employed is to cool the sensor.

Even for a relatively cheap $2000 astrocam with dual-stage peltier cooling, average dark current drops from around 5e- to 0.02e-, and better ones can be had for $4000 to $10,000 where dark current drops to as little as 0.01e- to 0.008e-. At 0.01e-, you release one electron worth of noise for every 100 electrons released by photons. Today, average read noise at high ISO is around 3e- or so, so cooling could gain us a fair amount of real-world high ISO sensitivity. Even at medium ISO settings, where dark current can still be as high as 5-10e-, could benefit from cooling. Extreme cooling could even be an option to reduce ISO 100 noise as well, albeit at a power cost.

On the notion of power consumption, that would certainly be a hurdle to overcome. Power cells would have to be far more efficient, and certainly hold more capacity, than even the most capable camera batteries of today. I suspect some kind of fuel cell technology would need to be employed to make extreme peltier cooling a reality for high ISO shooters. Fuel cell tech has come a long way recently, and I suspect at some point camera manufacturers will probably switch to them anyway. Thermoelectric cooling could be a user-selectable option as well, and the peltier could be activated automatically on demand if it is enabled so that it does not constantly draw power.

But is this realistic? How much would it add to the cost?

I've been familiar with cooling used on CCD cameras for astrophotography for a long time. Which is why I asked if it was realistic or feasible to do it in a DSLR.