In this post I have taken two Canon camera bodies that are quite different in many aspects and compared the images that each one produces.  The 1Dmk4 is currently Canon’s high-end sport and wildlife camera body (about to be replaced by the 1DX).  It is a fast, high-specification camera that sells for around the USD5000 mark.
The other camera in this comparison is the Canon 40D, which is a mid-level camera body that is no longer manufactured, having been superseded by two newer bodies in its price range, which was around USD1300 .
The 40D was announced in August 2007, while the 1Dmk4 was announced in October 2009.
At first glance, it may make no sense at all to compare any aspect of these two camera’s performance with one another.
However, if you dig a little deeper, the two camera’s share one important thing and that is their pixel pitch.  Both cameras have pixels of 5.7 microns in size.  They are also just over two years apart in age.  With this in mind, comparing image quality should provide a reasonable idea of how pixel efficiency has improved over time for Canon users.
Canon have tended to change pixel pitches often over the years, following a trend that saw each new model making use of smaller and smaller individual pixels.  This trend can make comparing advances in sensor and pixel efficiency a bit confusing when comparing different cameras, as sensor size, pixel pitch and efficiency can all be variables.
By making use of two cameras that have the same pixel pitch, the variables can mostly be eliminated.
The 1Dmk4 has an APS-H (27.9mm x 18.6mm) sensor which has 16 million pixels on it.
The 40D has an APS-C (22.2 x 14.8mm) sensor which has 10 million pixels on it.
The Mk4 has a larger sensor, but the size of the individual pixels is exactly the same as those on the 40D sensor at 5.7 microns.


Canon 1Dmk4, Cropped from 16 to 10mp, Iso 400.  Focal length 115mm.
Canon 40D, As shot at 10mp, Iso 400.  Focal length 115mm.
The image from the 1Dmk4 above was cropped from 16mp to 10mp, to exactly match the 10mp resolution of the 40D.  As the two cameras have identically-sized pixels, the resulting cropped image renders the subject at exactly the same size.  In other words, you can crop a 40D-sized image exactly out of a 1Dmk4 image by removing 6 megapixels.
I had to overexpose the 40D by just about a third of a stop in processing to get it to the same brightness level as the Mk4 image, but that should have little effect on noise and quality.
At this small size for web, the downsized images don’t look at all that different to my eye.  That is of course one drawback of having to work with web size constraints when viewing.
This also shows that it can be quite difficult to really evaluate an image that has been downsized to a small web preview.  To state this in another way, if your primary output for your images is going to be at small sizes on the internet, you might not need what the high-end cameras deliver.
Canon 1Dmk4, Cropped from 16 to 0.4mp, Iso 400.  Focal length 115mm.
Canon 40D, Cropped from 10 to 0.4mp, Iso 400.  Focal length 115mm.
In the next two images, I have cropped very heavily in order to show what is really going on with noise.  It is quite clear that the 1Dmk4 is showing a lot less noise in the image than the 40D image.  The difference is equivalent to about one-stop.  In other words the amount of noise on a 1Dmk4 image at iso 800 is similar to the amount showing on the 40D image at iso 400.
The 40D was not regarded as a particularly noisy camera in its time, and many of my published images were taken with the camera.  I usually tried to keep the Iso at 400 or below when possible.
The 1Dmk4 pixels benefit from just over two years of development, and it shows quite clearly in the solid improvement (I would estimate just less than one stop) when comparing raw image output with the older low to moderate iso settings.
Canon 1Dmk4, Cropped from 16 to 0.4mp, Iso 3200.  Focal length 115mm.
Canon 40D, Cropped from 10 to 0.4mp, Iso 3200.  Focal length 115mm.
With the next two images I kept the crop the same, to show detail, and shot both cameras at iso 3200.
Things change a little bit at high iso settings, with the difference between the two sensors becoming more pronounced.
The 1Dmk4 images are showing a fair amount of noise at this very heavy crop, but not nearly as much as the 40D.  The noise levels in the 40D image are very high, and are affecting the image quality.
At iso 3200 the difference between the two sensors is greater than when they are at iso 400.  This is revealing, in that it clearly shows how the newer sensor in the Mk4 is losing less quality with each higher iso setting than the older sensor.  In other words, the new sensor is falling off less steeply at higher iso settings.  At iso 3200 the 1Dmk4 is showing less noise than the 40D at iso 1600.  The difference is more than one stop, at these higher iso settings.
This comparison was conducted in very low light.  Shutter speeds when shooting at iso 400 were in the region of 1/10 second.  Unfortunately I did not have my toy zebra to photograph but made use of the furry baboon.  I purposefully positioned the subject so that the background would hold dark, unfocused areas to check for noise.  Noise is easiest to see when viewed parts of the frame that are uniformly low in contrast.
No noise reduction has been applied to these images.  I shot in raw, and converted the images in Adobe Camera Raw.  The only sharpening was the default capture sharpening setting of 25, applied evenly to both images.  I resized the images to 800×533 pixels in Photoshop and saved them at 190Kb.
In summary, the comparison shows that in just over two years, and using the same size pixel, technology advances between the 40D sensor and the 1Dmk4 sensor resulted in an improvement at low to mid-range iso settings in noise of just less than one stop, and at higher iso settings the difference was closer to two stops.
It is worth noting that the 1Dmk4 sensor benefitted from Canon’s gapless microlens technology in its design as well as processor advances.
Camera makers routinely claim massive improvements in low-light performance between successive models, but it is always important to note whether they are referring to heavily processed jpg images or raw output.  Usually the jpg images used as examples are so heavily processed by strong noise reduction applications that most fine detail is removed from them, and these claims may not stand up to real-world situations.
It is more realistic to compare noise on raw images such as those above.  There is of course much more to image quality than just noise.  On the other hand, noise is something that anyone who has photographed in low light will need to be aware of, and reducing noise is one of the most important aspects of improving overall quality.
There are significant improvements to be seen in the efficiency of the 5.7 micron pixels from the 2009-era sensor when compared to the 2007-era sensor.