Dynamic Range & Camera IQ

[Aglet]

I've got a 5d2 and a 7d where the banding is quite noticeable without even having to push shadows very much.

image I posted here:

www.canonrumors.com/forum/index.php?topic=9299.0

was shot raw with my 7D; ISO 100, 1/1000, f/8, all internal processing and ALO off. Firmware 1.2.5
PP in DPP used; +0.33 EV, contrast -2, highlight -5, shadow +5 to open up some "depth" in the foreground.

if shadow is left at 0, the banding is almost not discernable.
But any value above 0 the banding begins to appear in the dark area below the trees, this is in an area of RGB value of about 50, 35, 25, as processed, so not exactly near black and this is not much of a shadow push, frankly.

If I'd have had the lens I needed with me I could have shot this on my D5100 and the dark area would have been TOTALLY CLEAN even if I wanted to push it up to where the foreground was completely visible at lower midtones.

attached crop of banded shadow area at DPP settings describe above, (+5 shadow)
crop taken from below left of center

I've only tested 2 7D bodies, they were equally this bad. Some of you may have better ones, count yourself lucky.
The readout channel mismatch between the dual processors working on the 7D sensor's dual readout is likely responsible for the 8 pixel wide vertical stripes.

I'm reluctant to even try the v 2.0.3 firmware as it still may be buggy altho v2 firmware did improve my 5D2's banding considerably, if not enough to make me love it.

 

[spinworkxroy]

very interesting thread this is...these are tests i would never do on my own.
question is...how often do you have to PP your images so much to this extent? I've personally maybe only faces extreme DR scenarios like once or twice a year... maybe I'm just not shooting enough or I'm only shooting in ideal conditions..but for me..if the situation isn't ideal..I'd rather not shoot. maybe the Canon cameras suffer from all these mentioned problems..but unless you're saying every one of your shots are like this...I'm sure 99% of the time it performs great!

 

[nightbreath]

very interesting thread this is...these are tests i would never do on my own.
question is...how often do you have to PP your images so much to this extent? I've personally maybe only faces extreme DR scenarios like once or twice a year... maybe I'm just not shooting enough or I'm only shooting in ideal conditions..but for me..if the situation isn't ideal..I'd rather not shoot. maybe the Canon cameras suffer from all these mentioned problems..but unless you're saying every one of your shots are like this...I'm sure 99% of the time it performs great!

The paraggrapgh below contains thoughts straight out of my head that don't have any facts behind them. This is how I understand the camera IQ, so please correct me if I'm wrong.

For my work tonal range may be more important than dynamic range and can help me in getting all these nice colours in the image. While there's visible difference in DR (longitude of the RAW data, as I understand it – from black to white) of Canon vs. Sony sensors, I don't see any leaps in tonal range (latitude of the RAW data, number of tones) improvements:

• 5D Mark III -> 8.65 bits
• D700 -> 8.65 bits
• D800 -> 8.53 bits
• D7000 -> 8.5 bits
• 5D Mark II -> 8.48 bits
• 1Ds Mark III -> 8.44 bits
• 7D -> 7.89 bits

I'll be glad to hear everyone's thoughts on the topic

P.S. Tonal range data is taken from DXO screen measurements.

 

[nightbreath]

Any thoughts on tonal range affecting IQ? Or everyone is so obsessed with DR that there's no point in mentioning this?

 

[RLPhoto]

I'm a 550D user who has just yesterday ordered a 7D. I've never even looked through the viewfinder on a FF camera, let alone held one.

I do know my way around my camera, but on these boards I keep hearing people talking about Dynamic Range when comparing cameras or making wish lists of improvements.

Can somebody please point towards a good explanation of DR and also comparisons between similar cameras (e.g. D800 and 5D M2/3) so I can understand it better and how it impacts me? Also any explanation of camera IQ would be helpful.

Cheers.

For decades photographers have pursued higher and higher DR in camera's since the original daguerreotype. How much have we improved? Eh, Its hard to say in camera tech.

But,

In-out camera technique such as filters, timing, scouting, HDR (if you like that sorta thing), you can can easily photograph stuff that's far, far beyond anything a bare sensor can capture. Like 20 Stops.

So if you want better DR, Work on your technique to achieve so.

 

[ecka]

Image quality, dynamic range, high ISO noise, graphs, numbers, SNR (signal to noise ratio) ... all that makes a lot of sense when you know what it is all about. For me it's all about camera's per-pixel color reproduction performance. All the different tests are just for judging the same thing from different perspectives.
The easy answer for FF vs crop - pixel pitch. For example, 5D3 pixels (6.25 µm) are bigger than 7D pixels (4.3 µm), therefore they get more light and produce stronger signal (better SNR). 5D2 has even (slightly) bigger pixels, but the older sensor and processor technology makes it inferior to 5D3. SNR deficiency impacts false color (image noise basicaly) and false resolution of the "bayer filter" type sensor.

 

[Aglet]

Any thoughts on tonal range affecting IQ? Or everyone is so obsessed with DR that there's no point in mentioning this?

tonal range is going to be similar for most cameras because they have a similar overall signal to noise ratios around midtones, despite the differences in overall DR

if the SNR is improved, and thereby DR will likely improve along with it, then it's possible to use more digitizing bits to define a particular pixels signal level and that can lead to finer tonal gradations. (12 vs 14 vs 16 bit for example)

This is where medium format digital has an advantage with their big clean pixels and 16 bit digitizing, they seem to be able to produce better tonal gradients in the midtone and lower levels which make for smoother looking images than you get from smaller sensors. You can see this same effect to some extent by comparing FF 35mm digital with compact cameras.

however, most of that's rendered moot when final output is 8 bit-per color jpeg or similar 8 bit files used for printing.

until the final output is capable of utilizing more than 8 bits per color you're not likely to get more than 8 bits unless you interpolate downwards from higher sampling precision.

also, someone correct me if I'm wrong, but tonal range reduces as you move towards the darker areas simply because there are less bits there to work with (smaller numbers describing the intensity levels)
I think the tonal levels DxO mark measures are likely done around middle gray.

 

[dtaylor]

Mikael Risedal - the 7D yields nearly 11 stops in RAW, and a solid 10 in JPEG with HTP, and this is confirmed by multiple sources. Unless the testing was done in JPEG with HTP off, those graphs are simply wrong.

Side note - what drives me nuts about photographic conversations (and political conversations for that matter) is that people generally take sides. Once they have their side, they are hypercritical of any information that might support the other side, and completely uncritical of any information they deem to support their side. They also make the differences between the two sides seem larger than Mount Everest. This thread is degrading into that kind of nonsense.

There is no sign that Canon have invest 1 billion or more in new sensor lines to keep up with others.

You know darn well (or should know by now) that the difference in shadow noise between Canon and Sony sensors is due to a patent that is difficult to work around, and has nothing to do with the fab facilities or other tech on the sensor. As I recall from detailed discussions on dpreview, Canon is ahead on other points. But the way they read data off the sensor, and the banding that results, is what stands out at this time in (fanboy) Internet arguments.

Therefore it will be very interesting to se what the 46Mp roomers are about, and if it is the old 7d sensor stitched to a 24 x36mm then Canon has lost their credibility in my eyes

I would rather have the 7D sensor scaled to 46 MP than the D800 sensor. The differences in resolution and DR are small either way, but I would rather have the resolution. And I regularly push the DR in my photos.

 

[bdunbar79]

Rude? Are those your images? Posting obviously altered images seems to me to be rude.
I have been into photography for quite a few more years than you, started on digital earlier, and have owned or used many digital cameras, including two 5D Mark II's, two 5D Mark III's and a D800.
I've never seen the banding like that

How can you be earlier with SLR cameras? I have used scanning backs if that is what you mean to Hasselblads years earlier the first D-SLR came to the market ?

lower levels http://www.fredmiranda.com/5DIII-D800/index_controlled-tests.html

this is 5dmk3 and 5dmk2 in lover levels, Canon has smeared the 5dmk3 visual noise but it is the same structure
visible noise in a garage door from d7000 and 5dmk2 same exposure, same handling .

That has to be the most ridiculous thing ever. What picture would you EVER take that looks like either of them? Since the answer is obviously none, the test is ridiculous and meaningless. I've used both the 5D2 and 5D3 and even pushing photos, nothing has ever looked like that. You get a picutre back:

 

[dtaylor]

Now we come to the core, with a Sony sensor you can use the camera differently, you can underexpose to get highlights that are far above average gray...

Typical "defend my team" use of adjectives instead of mathematical precision: "far above". You can get about 1.5 additional stops of DR. Noticeable, can be nice to have at times, but not the end of the world. (Few people would consider a 10-15% gain "far above".)

While I'm at it...most of these comparisons expose the bodies in the same way. You would give the Canon a bit more exposure and recover the highlights in post. That's even noted in some of the comparisons, including one you linked.

 

[dtaylor]

tonal range is going to be similar for most cameras because they have a similar overall signal to noise ratios around midtones, despite the differences in overall DR
...
This is where medium format digital has an advantage with their big clean pixels and 16 bit digitizing, they seem to be able to produce better tonal gradients in the midtone and lower levels which make for smoother looking images than you get from smaller sensors. You can see this same effect to some extent by comparing FF 35mm digital with compact cameras.
...
however, most of that's rendered moot when final output is 8 bit-per color jpeg or similar 8 bit files used for printing.

Truth. There are tonal and color differences with big jumps (compact >> DSLR >> MF), but not so much with small jumps. And those differences don't necessarily make it to screen or print.

 

[jrista]

There is something so seriously wrong with that 5D II image. There is red banding and FPN throughout the entire image, even the highlights. There is NO WAY that image was properly exposed in the first place. I've seen that kind of banding in sample black-frame 5D II images, but only after pushing exposure by about 8 stops, or by opening the image in PS/ACR and using the levels tool to drop the white point to within a fraction of the black point. The 5D II's maximum saturation is 64600e- and its read noise at ISO 100 is 28e-. That is a ratio of 2307:1! No friggin way your going to see that kind of banding with a minor curve bump like you've demonstrated. That exposure would have to be underexposed by many stops to exhibit like that. I cry fowl!!

BTW, banding noise IS READ NOISE. Read noise is a bit too specific, as that generally refers to the noise introduced by the ADC. I tend to refer to electronic noise, which comes in a variety of forms, but is all very low relative to maximum signal (even when there is 28 electrons worth!) Its a bunch of bullhonkey when you mentioned before that banding is not taken into account when determining DR...absolutely it is, its simply that DR is computed as the AVERAGE of read noise to maximum saturation. Since it is the average of electronic noise, that that would mean your computing DR as the ratio between what would roughly be 14e- and 64600e-, a ratio of 4614:1 (or 11.something stops), so some electronic noise...such as banding...will show through in very deep shadows.

Sorry nothing wrong with this comparison, this is a example how you can handle the cameras , equal in terms of exposure and then adjust the images identical in Photoshop.

AND there is nothing as a proper exposure, some of you are living in a MYTH.
Banding and pattern noise from Canon are produced for instant that the 8, 16 readout locations not are in line with each other

[/quote]

The 5D II images appear to be very under exposed to me (and apparently, not just me). A "proper" exposure is one in which the image is NOT under exposed. And since the tests are comparing the recoverability aspects of both cameras, then you should be over exposing the 5D II image a bit, as Canon sensors have considerable highlight headroom. I bet you could over-expose the 5D II image by 2 stops, fully recover all the highlights, and have shadows that look nearly as good as the D800's.

 

[ecka]

Image quality, dynamic range, high ISO noise, graphs, numbers, SNR (signal to noise ratio) ... all that makes a lot of sense when you know what it is all about. For me it's all about camera's per-pixel color reproduction performance.

I care about images, not pixels. People that obsess with per-pixel image quality seems to be less interested in images that I am.

Per-pixel quality is academically interesting, but if you ever print your images or show them on a display, then that is the final indicator of quality. A 3 megapixel camera might have fantastic per-pixel quality (far better than my 7D), but who cares as long as my 7D takes better images??

-h

It's like saying "I care about milk, not cows" . I'm talking about pixel color accuracy and fake resolution. Have a look at some Sigma DP2 Merrill RAW samples (ISO 100) and you'll see what's the difference between the real 15 megapixel resolution and the fake 18 or 22 megapixel resolution from 7D or 5D3. Properly interpolated Merrill's RAW can give you a nice 30-40 megapixel Bayer-like image.
Please don't reply if you don't care.

 

[jrista]

There is something so seriously wrong with that 5D II image. There is red banding and FPN throughout the entire image, even the highlights. There is NO WAY that image was properly exposed in the first place. I've seen that kind of banding in sample black-frame 5D II images, but only after pushing exposure by about 8 stops, or by opening the image in PS/ACR and using the levels tool to drop the white point to within a fraction of the black point. The 5D II's maximum saturation is 64600e- and its read noise at ISO 100 is 28e-. That is a ratio of 2307:1! No friggin way your going to see that kind of banding with a minor curve bump like you've demonstrated. That exposure would have to be underexposed by many stops to exhibit like that. I cry fowl!!

BTW, banding noise IS READ NOISE. Read noise is a bit too specific, as that generally refers to the noise introduced by the ADC. I tend to refer to electronic noise, which comes in a variety of forms, but is all very low relative to maximum signal (even when there is 28 electrons worth!) Its a bunch of bullhonkey when you mentioned before that banding is not taken into account when determining DR...absolutely it is, its simply that DR is computed as the AVERAGE of read noise to maximum saturation. Since it is the average of electronic noise, that that would mean your computing DR as the ratio between what would roughly be 14e- and 64600e-, a ratio of 4614:1 (or 11.something stops), so some electronic noise...such as banding...will show through in very deep shadows.

Sorry nothing wrong with this comparison, this is a example how you can handle the cameras , equal in terms of exposure and then adjust the images identical in Photoshop.

AND there is nothing as a proper exposure, some of you are living in a MYTH.
Banding and pattern noise from Canon are produced for instant that the 8, 16 readout locations not are in line with each other

The 5D II images appear to be very under exposed to me (and apparently, not just me). A "proper" exposure is one in which the image is NOT under exposed. And since the tests are comparing the recoverability aspects of both cameras, then you should be over exposing the 5D II image a bit, as Canon sensors have considerable highlight headroom. I bet you could over-expose the 5D II image by 2 stops, fully recover all the highlights, and have shadows that look nearly as good as the D800's.

sorry but you are wrong, there is nothing like a proper exposure , you can expose as much as possible without clipping
and even then you can see the differences in the scene depending on how high/large dynamic range the scene has.
There is always a difference between 11 stops or 14 and 14 without pattern noise or banding
SO GO down in levels.


[/quote]

Well, I'm tired of swapping anecdotes with you. When you are ready to talk facts with proper reference and theory, I'll be waiting.

 

[dtaylor]

sorry but you are wrong, there is nothing like a proper exposure ,

:

There is most certainly a proper exposure if you wish to maximize image quality given the medium you are working with. This is true for B&W, slide, color neg, and each specific model of digital sensor. If Canon's strength is on the highlight side and Nikon's on the shadow side then you must tailor exposure to each one.

 

[ecka]

Image quality, dynamic range, high ISO noise, graphs, numbers, SNR (signal to noise ratio) ... all that makes a lot of sense when you know what it is all about. For me it's all about camera's per-pixel color reproduction performance.

I care about images, not pixels....

It's like saying "I care about milk, not cows" .

Yes, that is a very good analogy.

If you are a milk-drinker, the quality of the milk will probably affect the drinking experience. The color of the cow _might_ affect the drinking experience.

So why are you saying that you care more about the cow than the milk (to follow the analogy)?

-h

I suggest you start reading more carefully. I never said that I care more about the cow than the milk. I care about both actually. The thing is - if "cows" are fine then the "milk" is fine automatically, but not 'vice versa'.

 

[dtaylor]

It's like saying "I care about milk, not cows" . I'm talking about pixel color accuracy and fake resolution. Have a look at some Sigma DP2 Merrill RAW samples (ISO 100) and you'll see what's the difference between the real 15 megapixel resolution and the fake 18 or 22 megapixel resolution from 7D or 5D3.

Foveon is another area of great hyperbole in photography. An 18 MP Bayer sensor does not have "fake" resolution. There are 18 million sample points of luminance data. And a Foveon sensor does not have 3x its pixels in resolution. 15 MP Foveon has 15 million sample points of luminance data.

Foveon sensors do have more sample points of color data, and this results in better images then a mere count of MP would suggest. That can be seen and should not be denied. But it's not the night and day difference claimed by fans. Nor will they scale to match 30-40 MP Bayer images. This is the hyperbole. Go ahead and photograph a landscape with foliage using the Sigma and a D800e and see how the Sigma fares scaled to match.

With that out of the way, I wish Foveon would have had a chance with a larger and more aggressive company like Nikon or Canon. The technology is interesting. If it could have been pushed up in MP and ISO at a faster pace it might have had a much greater impact on the market. Again, it's not night and day. But it's certainly a strong edge in IQ.

 

[jrista]

It's like saying "I care about milk, not cows" . I'm talking about pixel color accuracy and fake resolution. Have a look at some Sigma DP2 Merrill RAW samples (ISO 100) and you'll see what's the difference between the real 15 megapixel resolution and the fake 18 or 22 megapixel resolution from 7D or 5D3.

Foveon is another area of great hyperbole in photography. An 18 MP Bayer sensor does not have "fake" resolution. There are 18 million sample points of luminance data. And a Foveon sensor does not have 3x its pixels in resolution. 15 MP Foveon has 15 million sample points of luminance data.

Foveon sensors do have more sample points of color data, and this results in better images then a mere count of MP would suggest. That can be seen and should not be denied. But it's not the night and day difference claimed by fans. Nor will they scale to match 30-40 MP Bayer images. This is the hyperbole. Go ahead and photograph a landscape with foliage using the Sigma and a D800e and see how the Sigma fares scaled to match.

Well said! ;D

With that out of the way, I wish Foveon would have had a chance with a larger and more aggressive company like Nikon or Canon. The technology is interesting. If it could have been pushed up in MP and ISO at a faster pace it might have had a much greater impact on the market. Again, it's not night and day. But it's certainly a strong edge in IQ.

Ditto. I was pretty excited when I saw the patent from Canon for a layered sensor design. I've looked at it a few times, and I'm not sure it compared to the current Foveon patens from Sigma, but I really hope/wish they would develop the technology further. I could totally go for a 22mp layered sensor.

 

[ecka]

It's like saying "I care about milk, not cows" . I'm talking about pixel color accuracy and fake resolution. Have a look at some Sigma DP2 Merrill RAW samples (ISO 100) and you'll see what's the difference between the real 15 megapixel resolution and the fake 18 or 22 megapixel resolution from 7D or 5D3.

Foveon is another area of great hyperbole in photography. An 18 MP Bayer sensor does not have "fake" resolution. There are 18 million sample points of luminance data. And a Foveon sensor does not have 3x its pixels in resolution. 15 MP Foveon has 15 million sample points of luminance data.

Foveon sensors do have more sample points of color data, and this results in better images then a mere count of MP would suggest. That can be seen and should not be denied. But it's not the night and day difference claimed by fans. Nor will they scale to match 30-40 MP Bayer images. This is the hyperbole. Go ahead and photograph a landscape with foliage using the Sigma and a D800e and see how the Sigma fares scaled to match.

With that out of the way, I wish Foveon would have had a chance with a larger and more aggressive company like Nikon or Canon. The technology is interesting. If it could have been pushed up in MP and ISO at a faster pace it might have had a much greater impact on the market. Again, it's not night and day. But it's certainly a strong edge in IQ.

I mentioned the Foveon X3 as a benchmark for Bayer sensor in terms of resolution (per-pixel color accuracy at low ISO). I'm not saying that Foveon X3 has 3x it's pixels in resolution. It's more like Foveon X3 delivers 95% of it's sensor resolution, while Bayer delivers only 30-80% of it's sensor resolution (depending on pixel size). Sigma cannot compete at ISO 400+, it's too noisy, but at ISO 100 there is a nigh and day difference compared to 18MP APS-C and even 22MP FF seems to be not quite as good. Google it . It does scale nicely up to 30-40MP, even if it is foliage landscape. However, D800E may have the edge, but it's not like night and day difference .

 

[jrista]

It's like saying "I care about milk, not cows" . I'm talking about pixel color accuracy and fake resolution. Have a look at some Sigma DP2 Merrill RAW samples (ISO 100) and you'll see what's the difference between the real 15 megapixel resolution and the fake 18 or 22 megapixel resolution from 7D or 5D3.

Foveon is another area of great hyperbole in photography. An 18 MP Bayer sensor does not have "fake" resolution. There are 18 million sample points of luminance data. And a Foveon sensor does not have 3x its pixels in resolution. 15 MP Foveon has 15 million sample points of luminance data.

Foveon sensors do have more sample points of color data, and this results in better images then a mere count of MP would suggest. That can be seen and should not be denied. But it's not the night and day difference claimed by fans. Nor will they scale to match 30-40 MP Bayer images. This is the hyperbole. Go ahead and photograph a landscape with foliage using the Sigma and a D800e and see how the Sigma fares scaled to match.

With that out of the way, I wish Foveon would have had a chance with a larger and more aggressive company like Nikon or Canon. The technology is interesting. If it could have been pushed up in MP and ISO at a faster pace it might have had a much greater impact on the market. Again, it's not night and day. But it's certainly a strong edge in IQ.

I mentioned the Foveon X3 as a benchmark for Bayer sensor in terms of resolution (per-pixel color accuracy at low ISO).

I think your falling into the same trap as most when comparing a Foveon with a Bayer. Bayer is only limited relative to Foveon in terms of color fidelity. A layered sensor design is capable of much greater color fidelity and accuracy because its capturing a full quantity of color information at every photosite. That also gives it another slight edge as it does not need a low-pass filter to eliminate color moire, since color moire doesn't exhibit. However Bayer sensors ARE detecting luminance data at every photosite, and there is no question they are capable of discerning a finer gradation of detail than a Foveon sensor...DESPITE the fact that their pixels are interpolated. I'm not sure an 18mp FF sensor is really going to be a significant edge, resolution wise, over a Foveon. But an 18mp APS-C sensor is going to resolve considerably more detail than a 15mp Foveon, and for that matter more detail than a 36.3mp sensor. Similarly, a 46.1mp FF sensor is going to be capable of the same resolving power as an 18mp APS-C.

The three-fold difference in luminance resolution and a far greater number of color pixels, several stops better ISO performance, and much greater spatial resolution, even when factoring in interpolation, gives a significant edge to Bayer in this case.