If you have read any of the site's other recent Canon EOS DSLR camera reviews, you will recognize the following color block test that clearly shows and compares sensor noise.
Read about the Camera ISO noise tests in the help section to learn more about the tests and how they are conducted. A key take-away from that page is that noise reduction is completely off unless otherwise specified.
Starting at the top, there are four mouseover bars of options from four cameras. The first thing you will likely notice is that the 40D and XSi samples were boosted by 1/3 stop in post processing. These two cameras produce an image 1/3 stop darker at the same camera settings. Various testing results showed reasonable consistency to this difference. The actual exposure samples from the 40D and XSi are available in the bottom set of mouseover bars for comparison. The difference is only 1/3 of a stop, but that is a helpful gain for the 50D and Rebel XS / 1000D.
Running a mouse straight down over the various camera model ISO setting labels will show you the difference in visible detail and the effective magnification of the sensor. As a generalization, higher resolution sensors show more detail than lower resolution ones. Focal length limited photographers (those needing more reach) will especially appreciate the extra detail the 50D delivers for them (there is more headroom for cropping).
As sensor density increases, so too does high ISO noise - unless improvements in the sensor and subsequent image processing are made. We have been seeing these improvements as the DSLR lines mature. What the above crops show - and what I've been seeing, is that the 50D has slightly more noise than the 40D at identical high ISO settings. Only a slight gain in noise from a sensor with 50% more pixels is still quite an accomplishment in my opinion.
Another fact made obvious in this comparison is that the 50D has two additional stops of ISO settings available - up to an incredible ISO 12800. Well, incredible until you look at the results from this ISO setting. They are ugly - I think there is more noise than subject at 12800. It looks to me like marketing wars between the competition are driving us places we don't want to be. To Canon's credit, ISO 6400 and ISO 12800 are extended settings and available only after being enabled in the proper custom function.
I actually had to use the extended ISOs while testing the camera. I shot a late-in-a-cloudy-day soccer game (no lights) with the 50D and a Canon EF 400mm f/2.8 L IS Lens. I started the game at ISO 6400, f/2.8 and quit mid-way into the second half at ISO 12800 and 1/640. Parents were turning car lights on to illuminate the game soon after I stopped - and the game was called just after that. The pics are poor from an image quality standpoint, but I have pics - and there was no alternative.
My noise observations thus far were made with noise reduction omitted from the equation. Noise reduction is an available in-camera (Standard/Weak/Strong/None) or post-processing option. Noise reduction makes a big difference in the images - both for the good and for the bad. The good is obviously that there is less noise in the images. The bad is that noise reduction can reduce image sharpness and visible detail. The goal is to find the best compromise. The "50D with NR" mouseover bar shows samples of Canon Digital Photo Pro (DPP) noise reduction settings ranging from 1,2 (Luminance, Chrominance) at ISO 100 through 8,14 at ISO 12800. I personally would rather have a little noise and retain the detail than remove all the noise and much of the detail with it. Fortunately, everyone can have their preference.
Another set of mouseover bars included above are "50D Sharpen=2" and "XS Sharpen=0" with the number indicating the DPP sharpening amount applied to the image. The Rebel XS / 1000D is so sharp at a "1" setting that moire can be seen in the image. The story behind the 50D samples at "2" starts with the ISO 12233 chart results from the 50D and Canon EF 200mm f/2.8 L II lens. Even at f/5.6, the results were not quite as sharp as I expected them to be. I next shot the ISO chart with the Canon EF-S 65mm F/2.8 USM Macro Lens. The results were very similar to the 100mm Macro. So, I shot the same test with the amazing Canon EF 200mm f/2 L IS USM Lens. While these results were slightly better (to be expected), they were not strikingly better. The test chart is very harsh on cameras and lenses, but does show even small differences. Canon support has not been able to get me the reason for the slight softness at the pixel level, but because it is present in even the best lenses, I'm going to *guess* that the anti-alias filter may be a little stronger relative to the individual pixels on this sensor.
While not quite as sharp at the pixel level, the 50D is still pulling more detail out of the scene. The fabric example above show this. As you drag your mouse pointer up and down over the 50D and 40D labels, watch the tiny threads appear in the 50D image that are not present in the 40D image.
Downsizing an image makes it sharper (if proper sharpening is applied) and reduces the visibility of noise. Reducing the 50D image to the pixel dimensions of the 40D results in very similar noise levels and the 50D retains better sharpness and details. The 50D has two small RAW image format options (sRAW1 and sRAW2) available (reduced size JPEGs are of course available). With no noise reduction applied, shooting in the 3267 x 2178 sRAW1 format results in much less visible noise - ISO 3200 results actually look quite good and ISO 6400 becomes usable. I don't find sRAW1 images to be any sharper than RAW, but sRAW2 (2376 x 1584) images are very sharp. When all 3 RAW formats are processed in DPP to the 3267 x 2178 sRAW1 size, the full-sized RAW files produce sharper details with a more fine-grained noise. Applying noise reduction on the reduced RAW image until it is similar in sharpness to the sRAW1 image makes both images very similar in noise levels. So, the primary gain I see for using sRAW is in reduced file size - 24,680 KB vs. 15,571 KB vs. 11,647 KB. in one ISO 3200 comparison example.
At higher ISO settings, a detailed subject is going to show less noise than a smooth subject with no noise reduction applied - the details hide the noise better than evenly colored surfaces. As discussed before, noise reduction reduces the visible details - but smoothly colored surfaces do not have much detail in the first place. So, noise reduction can be more heavily applied to non-detailed images. Following is a comparison example with color blocks. Notice that at ISO 3200, color blocks themselves look dramatically better with noise reduction applied.
The comparison examples above were shot identically to the prior comparison examples with the exception that the Canon EF 70-200mm f/4 L IS USM Lens (at 100mm) was used. Identical exposures were used.
I spent a lot of my 50D review time pixel peeping between these 4 bodies. After using the 50D for a week in Acadia National Park (Maine, USA), I had already proven to myself that the camera functions very well. But I found the pixel-level image sharpness from this camera very difficult to evaluate - more difficult than any Canon DSLR I've reviewed to date. Some shots seem very good and others slightly softer than I think they should be. I use full-frame bodies for most of my work, but need a 1.6x FOVCF body for reviewing EF-S (and similar) lenses. Image sharpness is an important lens feature to evaluate and at this time, I must choose between the 40D and 50D for my future review body. Image quality is the most important criteria for my decision. Because of the extra detail the 50D is pulling in, my current plan is to go with Canon EOS 50D for this work.
Overall 50D image quality including the color it delivers is excellent. But, I expected that. The auto white balance in the first fabric crop was slightly different from the 40D, but I did not notice any big changes overall.
Back to the Canon EOS 50D Review.
Bringing you this site is my full time job (typically 60-80 hours per week). Thus, I depend solely on the commissions received from you using the links on this site to make any purchase. I am grateful for your support! - Bryan