Another 7D ISO Noise Test

I finally got around to doing the analysis on the test shots I did with my 7D and 20D when it arrived and thought others may be interested.

I'll just post the results discussion and conclusion plus the principle noise graphs. More details and the 100% crops can found at the test report page by clicking here.

For a meaningful comparison it is necessary to compare noise and sharpness and try and keep one parameter nearly equivalent. This test is designed to check both.

The EF 100mm f2.8 macro USM lens at f8 was used in all cases. All 7D exposures have the same focus setting as do all the 20D exposures.

Two sets of results are produced, No Noise Reduction where the sliders for luminance and chrominance noise reduction and DPP sharpening are set to zero.

The second set Default Noise Reduction, uses the DPP default luminance and chrominance noise reduction settings but still sets the DPP sharpening to zero.

The 20D RAW, 7D RAW and 7D MRAW are then also down sampled to the 7D SRAW size (2592 X 1728 ) using the Photoshop "bicubic sharper" algorithm. This provides a common resolution size for comparison that is close to a minimal print resolution (230 dpi on A4).

No Noise Reduction Results.

Although this gives the sharpest results the noise performance of the 20D and 7D is essentially the same over their common ISO ranges.

Downsizing by using MRAW or SRAW or by using the bicubic sharper algorithm on the RAWs reduces the noise levels a little. These trends are fairly clear for the luminance noise, but less so for the chrominance, however the chrominance noise is fairly well clustered.

For the RAWs the sharpness degrades towards the higher ISOs as might be expected. Using smaller RAW formats or downsampling compresses together the resolution traces and reduces the effect of the high ISO, again as might be expected.

Default Noise Reduction Results

Using the default DPP noise reduction settings reduces the 7D files sharpness slightly, but has a dramatic detrimental effect on the 20D file MTF 50% sharpness that is apparent even after downsampling.

Despite this, the reduction in 7D luminance and chrominance noise is very significant compared to the 20D [The default DPP settings for the 20D do not include any chrominance NR!!]. The luminance advantage is about two stops and the chrominance advantage is about 3 stops at the maximum ISO settings.

Again, downsampling or lower resolution RAWs reduces noise levels, sometimes not by much, although for the SRAW this reduction is more significant.

Noise Performance at Comparable Sharpness Levels

The sharpness of the 20D RAW image is comparable or slightly lower than the 7D RAW image when the 20D has no noise reduction and the 7D has the default noise reduction.

On this basis, at 3200 ISO, the luminance RMS noise on the 7D is about 3 and the 20D is about 8. For the same sort of noise level the 20D would need to be around 400-800 ISO, so a 2-3 stop advantage.

Similarly, for the chrominance noise at 3200 ISO the 7D achieves an RMS noise level of about 1.5 and the 20D about 5.5. To achieve 1.5 the 20D would need to be at 100 ISO and is still not quite as good, so a difference of 5 stops or more.

Due to an oversight the plot images are a bit large so will have to post them as links:

No Noise Reduction Results Plots

http://www.zen20934.ze​n.co.uk …ots/Black%20lum​inance.jpg

http://www.zen20934.ze​n.co.uk …k%20b-g%20chrominance.jpg

Default Noise Reduction Results Plots

http://www.zen20934.ze​n.co.uk …ots/Black%20lum​inance.jpg

http://www.zen20934.ze​n.co.uk …k%20b-g%20chrominance.jpg

Lotta data here to pound through. Gonna take a few readings to digest it all.

Appreciate the effort. But this one's way over my head. Lost me by 1st paragraph I think. More so because I'm rather impatient and need to see photos, not just plotted charts. Dots on a graph are just dots on a graph. Having the actual image behind how the dots are derived works better for *me*. (Certainly doesn't mean that others won't greatly appreciate and understand all the data you put up.)

Lester, I for one, appreciate the time you put in, providing the graphs and conclusion. I actually like what you have done, when it comes from a trusted source like you. I dont know if the loss of detail in dpp for 20D images is due to poor algorithm, since with no processing the raws are comparable.

The technical stuff is over my head but the summary in the post is easy to understand. Thanks for the great information.

By "default" noise reduction on the 7D, I assume that you mean that the NR is set to "Standard" or it's intermediate level.

Interesting results - thanks for performing this test.

It was a little difficult differentiating the different colored lines on the chart on this computer (I'm not at home), but it looks like the 7D with "Standard" NR provides both better resolution and much lower noise at high ISO settings.

Thanks all, I hope it is helpful.

Julian I appreciate your point. Of course shots of test charts and lots of nice graphs is always just a starting point. The idea is after to make photographs. For me this gives an idea of how much I can push the ISO value in practical situations compared to my old camera.

Tom, yes indeed, that is my conclusion as long as you use some NR on the 7D. The default NR is just what DPP set by default . To be honest I only shoot RAW and this is the first time I have used DPP since I first got the 20D 5 years ago, so I am not too sure of the settings on the camera or DPP for NR. I just recorded them in a spreadsheet, you will find a link to a jpeg of this on the test report page. I am sure there are a lot of prametric studies one could do to optomise this, I was kind of hoping the defaults represented the conclusions of such work by Canon, but perhaps I am too optomistic.

I guess it is blindingly obvious that a higher resolution sensor will take more NR without loss of absolute resolution, but I always find it best to confirm ones instincts with hard data.

I don't have my camera handy right now, but I was under the impression that changing the noise reduction in camera also changed the settings applied in DPP on a RAW image. Of course, in RAW, you can change that back which you did in your test.

I'll have to check tonight when I get back home.

Tom W wrote in post #8849432
I don't have my camera handy right now, but I was under the impression that changing the noise reduction in camera also changed the settings applied in DPP on a RAW image. Of course, in RAW, you can change that back which you did in your test.

I'll have to check tonight when I get back home.

I am sure this is right Tom, I never shoot jpg and normally use ACR so I don't take much notice of these camera settings.

Tom W wrote in post #8849432
I don't have my camera handy right now, but I was under the impression that changing the noise reduction in camera also changed the settings applied in DPP on a RAW image. Of course, in RAW, you can change that back which you did in your test.

I'll have to check tonight when I get back home.

I believe what DPP does is based on one of th preferance screens. You can have it set to use the camera defaults (the default mode) or you can manually set it to use a preset. I've set my DPP to always use no noise as a default.

I'm sure plenty of people will appreciate this information and I certainly appreciate all the effort that's gone into producing this, but…

I agree with Julian…I need to see actual photos; graphs tell me nothing about how a camera reacts to real life situations.

xxloverxx wrote in post #8850900
I'm sure plenty of people will appreciate this information and I certainly appreciate all the effort that's gone into producing this, but…

I agree with Julian…I need to see actual photos; graphs tell me nothing about how a camera reacts to real life situations.

Plenty of posts with real pics as well as full-sized OOC JPGs to review, just need to search a bit (search term = 'penguins of madagascar' for example)

Yes that is right lots of pics about.

Conversely they tell me almost nothing because of the large number of variables in real world shots, I like to start with hard facts then move on.

Lester Wareham wrote in post #8851139
... I like to start with hard facts then move on.

I agree. Very little can be gleaned from real world shots. There are just too many variables that come into play.

Thanks, Lester. One thing that does not show up in statistics but is very important is pattern noise. On the 7D, pattern noise at ISO 800+ is completely gone, whereas on the 20D it is terrible. The patterns may only affect the std dev. and charts by a few percent, but human brains make the lines stand out like a sore thumb. Just FYI.