Are new High ISO cameras just blurring images?

[Robert Park]

I had the opportunity to try out a 7D twice. I was in Yellowstone 3 weeks ago and got to compare a friends 7D to my 5D2 and his 1DIIn. I felt the 7D had great rez but seemed to put a blurr in shadow areas on dark animals and shaded foliage. We shot raw at 100, 800 and 1600.
Granted at 100 it all looked quite good. The 5D2 and even the 1DIIn looked like they retained detail better at 800.
At 1600 I personally didn't like the noise from anything but I really don't like noise at all.
However the 7D surprised me with it's loss of detail in those dark low contrast animal areas at 1600.
I have heard such good praise of the 7D I decided to try again so I just got back from a test with a loaner from my local shop. I am getting similar results. Granted I am pixel peeping but I want to make quality enlargements and pixel level peeping is the revealing factor on how a 24x36 will look.

Is it me or do the new high ISO cameras get that iso rating by blurring the raw file or am I somehow not processing the 7D correctly? That old 1DIIn actually looked wonderful at 800 and that shocks me.

[Fabs Forns]

You are comparing a pro camera with a prosumer camera. Nor fair. And yes, the 7D tends to be noisier than the other two models. Keep in mind that detail could depend on the subject. I find that feather detail goes faster than let's say bugs, with no feathers, so I could generally go higher ISO in macro than I would for birds.
BTW, we go by real names in the forum, so please drop Alfred Forns a PM so he can fix yours.
Big welcome to BPN!

[Desmond Chan]

I had the opportunity to try out a 7D twice. I was in Yellowstone 3 weeks ago and got to compare a friends 7D to my 5D2 and his 1DIIn. I felt the 7D had great rez but seemed to put a blurr in shadow areas on dark animals and shaded foliage. ...

I think the title should have been more specific.

[arash_hazeghi]

you cannot blur a "RAW" file. You must be applying different amounts of noise reduction in your RAW converter, most third party RAW conversant apply aggressive and lousy NR to 7D files by default which makes it worse than what it actually is. Obviously the 7D is way noisier than 5D2, it has a stronger AA filter too and its small pixels put more burden on the optics so its files look softer and less crisp compared to 5D2 when viewed at 100% at all ISOs, but 7D does not "blur" its RAW files either. To get the best results from 7d don't use 3rd party converters, use Canon DPP and turn off the luminance noise reduction function, use the chroma NR slider to get rid of color noise that appear as you increase the ISO. Whether the pixel level sharpness and noise of 7D files is acceptable to you or not is a matter of personal preference, some people will accept and praise it and some will reject it-like many other things it depends on how critical you are.

Here is a 100% pixel view of 7D at ISO 1600 processed with DPP software with optimized sharpening, this is pretty much as good as it gets

http://www.stanford.edu/~ahazeghi/Photos/examples/crop_heron.jpg

here is the FF image processed with a technique I have described in ETL

http://www.stanford.edu/~ahazeghi/Photos/birds/GBH%20portrait%20low.jpg

[Robert Park]

Arash

Thanks for the tip. We used ACR to process the images and possibly that has a bit to do with it. I do remember using DPP a year ago and noticing it was far less noisy and better color density than ACR.The noise structure in your image is much finer than what we saw on the 7D. We saw chunky blobs and poor definition in areas like what you illustrate, it was much more uneven. Even my 5D seems coarser than that from ACR. I think Canon definitely knows their secret sauce better than Adobe.

BTW that is an excellent illustration and testimony to your expertise IMHO

[Roger Clark]

Robert,

The noise you see in the images is related to the pixel size. For a given lens, f/ratio and exposure time, the camera with the smaller pixels will have less light per pixel. It is like cutting up a pie: if you want more pieces, each one must be smaller. The signal-to-noise ratio digital camera images in all but the deepest shadows is proportional to the square root of light recorded in each pixel. So in a camera, like the 7D, where the pixels are spaced every 4.3 microns, each pixel collects less lightl than the pixels in a 1DIIn with 8.2 micron spacing. So you trade a high signal (and signal-to-noise ratio) for higher spatial resolution. When the signal is low (what you have when working at high ISO) then many people perceive that noise as detrimental to the image quality and prefer a lower resolution better signal-to-noise ratio image (e.g. 7D versus 5DII). My personal view is that the higher spatial resolution of the 7D is great at relatively low ISO and at high ISO I prefer a camera with larger pixels.

If you want to delve more into this subject see my web page:
http://www.clarkvision.com/articles/digital.sensor.performance.summary/

Roger

[Robert Park]

Roger
Wow what a read. Great pages on your site too.
Thanks
Robert

[John Chardine]

Robert- You don't say if you were shooting RAW or JPEGS. What high-ISO noise reduction setting was used? Am I wrong in thinking that this in-camera NR works with RAW as well?

Arash- could you elaborate on "optimised sharpening".

[John Chardine]

Arash- could you provide a link to the ETL thread you mention. Thanks.

[Roger Clark]

Robert- You don't say if you were shooting RAW or JPEGS. What high-ISO noise reduction setting was used? Am I wrong in thinking that this in-camera NR works with RAW as well?


In camera noise reduction does not appear to work on the raw data in Canon cameras.

Roger

[Robert Park]

I think my issues might have been a set of compound problems. I have researched the 7D and think that my standard practice of being comfortable underexposing on the 5Dmk2 does not work on the 7D.
I also noticed that I felt the 7D was overexposing images as the standard exposure had no real density in shadows so I exposed for a histogram similar to what experience had shown to work in the 5Dmk2.
I have seen that the dynamic range is much more limited on the 7D and that people have bad experiences with underexposing that camera. I also was using ACR which seems to make a lot more noise/uneven patterns. The 7D is a different beast and probably needs a bit more time to master than just grabbing it and shooting it like another Canon.

[Emil Martinec]

The 7D has a number of 'features' that contribute to noise:

1) The two green filters in the Bayer RGGB array have slightly different spectral response (one is more "yellow"). If a raw converter is not careful about treating this, it leads to greater noise. Even if it treats it properly, the methods I have seen for dealing with it can rob the image of some pixel-level detail at lower contrast levels.

2) The 7D, like the 5D2, has a lot of pattern noise (perhaps due to underspeccing the readout electronics for the data throughput needed with high MP counts). The line noise also contributes to overall noise, and robs from the useful DR.

3) As Roger says, smaller pixels have individually lower S/N ratio for a given level of exposure. However, this doesn't mean that the image is more noisy than another image taken with a sensor of the same size having fewer pixels. Differences in noise for a fixed size image will be negligible; only if you zoom in to 100% will you see differences, because you are magnifying the image more when the sensor has smaller pixels, and noise is greater at finer scales. Image S/N at a fixed scale doesn't change much with pixel size, though the effect is greatest at high ISO where the noise from camera electronics is a larger component of the overall noise.

Because of (1) and (2), the 7D is rather sensitive to which raw converter you use. You said you used ACR, but which one? The ACR6 in CS5 is a much better converter than the ACR5 in CS4.

[Robert Park]

Emil
When I was in Yellowstone we used my friends laptop and I know he has CS4 however I have CS5 and LR3. I am thinking I might just be particularly sensitive to noise.
Do keep in mind that I am primarily a LF film shooter and am not accustomed to much noise. I chose the 5D2 for low noise and rez.

I also am in need of making big prints that wont look bad next to a LF film 30x40. Granted we are 2 totally different realms but I have made nice enlargements from far less
rez than a 5x7 trans or 21mp has. I have made 30x45 enlargements for local photographers who use the original 12mp 5D that are really quite good and a large part of it was from lack of noise.
I am really quite new to the digital wildlife arena and have made my living from large prints from LF film so I might seem a bit ignorant on some of these fine details that you people have
had the time and experience to hash through.

I guess I was expecting more from the 7D with all the hype I have heard that it broke bounds and yada yada yada...

Anyhow this is my staple
http://www.plateaulight.com/Personal/Bob-Entrada.jpg

[John Chardine]

Robert- I think you would be pleased with the results from the new 1D mkIV. Roger has analysed its performance and it is "off the scale" in a lot of measures related to IQ (hope I have this right Roger!). It is certainly the best camera I have ever owned, and it better be for the price Canon asks!

[Robert Park]

John
I have begun the lengthy task of digesting his site and was coming to that conclusion or just using my 5D2 to it's potential until i'm absolutely sure.
Roger has created prints from LF film also and I am willing to bet he know a thing or two about quality images.
Thanks
Robert

[arash_hazeghi]

Arash- could you provide a link to the ETL thread you mention. Thanks.

http://www.birdphotographers.net/forums/showthread.php?54826-Grains-in-7D-RAW-files

[Roger Clark]

All,
As one may surmise from the discussion so, far the situation with various sensors is complex. It boils down to the fact that no one camera is the best tool.

I can cite several examples. For example in focal length limited situations, like that commonly found in bird photography, one strives to get as many pixels on the bird with the longest focal length lens one has. The more pixels, the larger the enlargement, or the smaller pixels the more one might crop and maintain a lot of pixels on the subject. Thus, one magnifies those pixels and noise can be more apparent. This situation is described in section:
http://www.clarkvision.com/articles/digital.sensor.performance.summary/index.html#FLLAIQMAX
and the smallest pixels will generally give the best results for a given lens.

Emil points out that if you make the same size print, e.g. 8x12 from the full sensor, then the noise apparent in the image will appear about the same regardless of pixel size. For example, 8x12 prints from a 10 megapixel 40D versus 8x12 print from the 18 megapixel 7D would both show about the same noise. I agree. But more often one wants to enlarge the little bird in the frame more with the 7D than the 40D image because you have more pixels. Then the noise difference becomes apparent.

For those who want to make larger and larger prints (e.g. present and former large format photographers) one wants to push the limits. I commonly made 30x40 inch prints from 4x5 velvia drum scanned. No current digital camera can match the quality in a single frame of 4x5 velvia (well perhaps some of those 4x5 scanning backs could). So if one made a large print from a 40D, 7D and 5D2, the 5D2 would get the closest. This situation is quantified in my AIQ plot, Figure 13 at:
http://www.clarkvision.com/articles/digital.sensor.performance.summary/
where we see a large spread between sensors.

If you want to make a given print size of a subject in a focal length limited situation, e.g. a bird appearing 7 inches tall on an 8x12 inch print, and the bird is small in the frame, the situation is more complex. The sensor characteristics for this situation is shown in figure 14 at:
http://www.clarkvision.com/articles/digital.sensor.performance.summary/

So which camera makes a better image for a given situation will vary with the subject, available lenses, light level, and intended enlargement. No one camera is tops in all conditions. Also important is apparent signal-to-noise ratio (S/N). If S/N is higher, getting it higher will not help perceived image quality. If S/N is low people do not like the image quality, even if it has more detail (and where that trade point is varies between people).

For example, if you want to do frame filling subjects so position or focal length is not limited, the a full frame sensor camera with a lot of large pixels (for good S/N) does a very good job.

But if you are focal length limited photographing a small bird at great distance, the camera above will not do a good a job as a camera with smaller pixels, assuming good light.

So, in the Canon line (because Robert uses Canon), here is what I would use:
Landscape and big enlargements: 5D2 or 1DsIII (both full frame 21 megapixels)

Wildlife in focal length limited situations: 7D (tiny pixels for finer detail) or with enough money, 1DIV (not as small of pixels, but you can use 2x TCs on f/4 supertelephotos and still have AF.

As a large format photographer, I am also pushing limits with large prints. I have found that digital mosaics can give greater than 4x5 film quality. See, for example:
http://www.clarkvision.com/articles/large_mosaics/

Then another subject is low light, high ISO photography, including night and astrophotography. In that case, larger pixels do help, if the read noise is low (which these days is really good on most digital cameras). In this arena, I would not choose a 7D, but would use a 5D2 or a 1DIV. (Note, Nikon has equivalent great cameras in these areas too.)

Given all of this, one can still make great images with any of the cameras. I've used my 5D2 for wildlife and bird photography with great success, as have many others on this forum. Similarly with the 7D and other cameras.

As I've stated in another thread, it seems the manufacturers are putting just enough different features in cameras to maximize the desire for each one. For example, the ideal setup would be (for Canon): a 5D2, 7D and 1DIV, with the luck to have the right lens on the right camera for the right situation, and the strength to carry all that gear:):cheers:
Oh well. Life is a compromise.

Roger

[Emil Martinec]

I also am in need of making big prints that wont look bad next to a LF film 30x40. Granted we are 2 totally different realms but I have made nice enlargements from far less
rez than a 5x7 trans or 21mp has. I have made 30x45 enlargements for local photographers who use the original 12mp 5D that are really quite good and a large part of it was from lack of noise.
I am really quite new to the digital wildlife arena and have made my living from large prints from LF film so I might seem a bit ignorant on some of these fine details that you people have
had the time and experience to hash through.

I guess I was expecting more from the 7D with all the hype I have heard that it broke bounds and yada yada yada...

Anyhow this is my staple [image]


Where the 7D broke bounds has to do with the electronic noise generated by the camera, per pixel, at high ISO. Not something terribly relevant for your sort of photography. But generally, lots of hype comes with each camera generation, yet one is not going to evade basic physical bounds that dictate how much noise will be present in the image.

I agree with what Roger is saying. What you are observing for the most part has less to do with resolution and most to do with sensor size. With large format film one has a "sensor" with 15x the area of 35mm, and 35mm (eg 5D2) has 2.5x the area of the 1.6x crop of the 7D. The number of photons gathered for a given exposure goes with the area, and the S/N goes with the sqrt of the area, or the crop factor; all this assuming that the digital sensor and film are not too different in their efficiency in capturing incident light. If you took a 4x5.3 crop of your 30x40 LF print and enlarged it back up to 30x40, it would (a) come from as many photons as the 7D, and (b) have as much or more noise. Put another way, a 7D shot at ISO 100 has the same noise as full frame shot at ISO 250, which in turn has the same noise as 4x5 film shot at ISO 4000. Or, the photon gathering capacity of large format film is as much more than typical DSLR sensors as the DSLR sensor has compared to typical P&S cameras. So, just as expecting a P&S to hold its own against a DSLR is asking too much, so is asking the 7D to do the same side-by-side with LF.

All that in situations where one can use the entire frame. If one is focal length limited, then one is cropping the frame to keep the subject a desired size in the final image, and so one should factor in the percentage of crop in the calculation of light-gathering ability, and therefore noise. A 1.6x crop of the 5D2 would look about the same for noise as the 7D when viewed at the same print size, it would just have about 1/3 less resolution, or about half the number of pixels on the subject.

BTW, much of the noise difference goes away if you ask the various formats to take the same image in terms of shutter speed, framing and DoF. Since the larger format has to be stopped down more to achieve the same DoF, for a given shutter speed (and adjusting the focal length according to the crop factor so that the framing is fixed, ie assuming one is not focal length limited) the total number of photons being gathered is the same when the DoF is held fixed, so the images will look the same as far as noise (and of course, DoF, framing etc). So the advantages of the larger format comes from a tradeoff of DoF for increase light-gathering, assuming one has the focal length lens that one can do the framing in-camera rather than cropping the image after the fact.

[Doug Brown]

I have researched the 7D and think that my standard practice of being comfortable underexposing on the 5Dmk2 does not work on the 7D.


Underexposure with the 7D is not a good idea IMO, for the reasons outlined above. There's not a lot of detail in the darks (but there is plenty of noise) when you try to bring them up in post.

[Roger Clark]

I agree with what Roger is saying. What you are observing for the most part has less to do with resolution and most to do with sensor size. With large format film one has a "sensor" with 15x the area of 35mm, and 35mm (eg 5D2) has 2.5x the area of the 1.6x crop of the 7D. The number of photons gathered for a given exposure goes with the area, and the S/N goes with the sqrt of the area, or the crop factor; all this assuming that the digital sensor and film are not too different in their efficiency in capturing incident light. If you took a 4x5.3 crop of your 30x40 LF print and enlarged it back up to 30x40, it would (a) come from as many photons as the 7D, and (b) have as much or more noise. Put another way, a 7D shot at ISO 100 has the same noise as full frame shot at ISO 250, which in turn has the same noise as 4x5 film shot at ISO 4000. Or, the photon gathering capacity of large format film is as much more than typical DSLR sensors as the DSLR sensor has compared to typical P&S cameras. So, just as expecting a P&S to hold its own against a DSLR is asking too much, so is asking the 7D to do the same side-by-side with LF.

Emil,
A couple of points here. Film is less efficient than digital sensors, on the order of 10x. Thus your comparison should be more like ISO 400 4x5 film. But in practice there is also human perception in the equation. Some people will choose a slightly noisier image with higher spatial detail over lower noise lower resolution images. For example film has pretty low S/N but people made beautiful images from film (and some still do).




BTW, much of the noise difference goes away if you ask the various formats to take the same image in terms of shutter speed, framing and DoF.

But that is not reality, is it? For example, large format typically needs f/22, f/45, even f/64 to get the same depth of field as 35mm cameras, so one uses a longer exposure time.

In the digital world, for landscape photography, I would recommend the landscape photographer who wants to compete with larger format to get the largest sensor, high megapixel camera and do mosaics whenever possible, e.g. 3x3 mosaics with a 20 megapixel camera (with about 20% overlap between frames) can do very well against 4x5 film. Larger mosaics can beat 4x5 and even 8x10 film. And one trick I have learned is to let the focus move between frames, then I can use a faster f/ratio and faster exposure time and get images in situations where I could not get a 4x5 film image.

Roger

[Robert Park]

Roger
Oh yes. I am fully aware of the possibilities of mosaics :) My tripod has not had a ball head for a year as I leave the gimbal on all the time.
The gimbal works great with a RRS slider for a full 3D very close to nodal setup and I can always get the big lens out an mounted and i'm ready to go anytime.
I used to split hairs with the exact nodal point position but software is so good i just get close enough and concentrate on shooting.

[Emil Martinec]

Emil,
A couple of points here. Film is less efficient than digital sensors, on the order of 10x.
Roger

OK, good to know, I didn't realize film was that bad in terms of efficiency. Thanks.

[Hal Everett]

Arsash's methodology for dealing with noise from the 7D is IMHO, the best way by far to process any 7D raw file with an ISO above 400. I was about to toss my 7D before I read it.

However, where I live, and especially this darkest and grayest of years, that is not enough. I take more shots at ISO 1600 than any other ISO setting. The 1D IV is far better, but I am looking with envy at the Nikon D3S. My trip to YNP in late May would have been a total loss without the 1D IV(we had 3 hours of sunshine in 10 days).

If low light, high ISO shooting is your game, the 7D should probably not be your first choice. But, I love it when the light is good..