If downsampling from a higher resolution sensor [say that of the 50D] can indeed match the resolution and noise level combination of a lower resolution sensor alternative [say the 40D], then the higher resolution options is clearly superior overall:
A. in some situations like at sufficiently low ISO speed, the noise levels of the higher resolution option are completely acceptable, so no downsampling is needed and you get more resolution than with the lower resolution sensor.
B. in other situations, the option of downsampling means that you are never worse off than with the lower resolution sensor.
So the higher resolution sensor sometimes wins, sometimes draws, and never loses. Such downsampling/NR options are a lot like being able to switch between a higher resolution lower ISO film/sensor and a lower resolution higher ISO film/sensor according to the situation.
BJL,
There's an excellent explanation of this process on the new DXOMark website under the 'Insights' heading. The DXO team have mapped the S/N of 30 DSLR models produced since 2003. Surprisingly, S/N is getting worse. There's a downward trend as pixels get smaller. However, when images are normalised to a specified size, an opposite trend can be seen. Signal-to-noise with respect to the entire image is steadily increasing.
I do not feel much need for 15MP final cropped images, but like the idea of cropping latitude for wild-life photography, to get a well-composed 8 or 10 MP crop image of an unpredictably moving subject.
Perhaps you would change your mind if you had a 50" (or bigger) plasma display .
We seem to be obsessed with resolution matters in photography because people have a habit of approaching as close as possible to a print and peering at the detail. If the detail is lacking on close inspection, then in some way the photo seems to fail a reality test. We are perhaps disappointed in some way. In the real world, the closer you look at an object, the more detail you see.
My own tests have shown that, using a cheap standard lens at apertures from F4 to F16, the 50D will always produce more detail than the 40D, provided the detail exists in the target and provided one is prepared to enlarge the image sufficiently for that extra detail to be revealed. The problem is, unless one is in the habit of making 4ftx6ft prints and larger, such additional detail is often too insignificant to be relevant. Also, if one
were in the habit of making 4ftx6ft prints, one would probably want to use a 50mp DB to make such increased resolution and detail worthwhile.
The HD plasma screen comes to the rescue. My 50" plasma display is as high as my large format Epson 7600 is wide. I cannot make a larger print, in the 16:9 aspect ratio, than this plasma screen can display. Despite the resolution of the plasma display being only 1920x1080, from a comfortable viewing distance of anywhere from 8ft to 12ft, I see as much detail as I see on a print of the same size from the same distance.
Of course, if I approach the print up close, I can marvel at the texture of the moss on the rocks. If I get close to the plasma screen, I begin to see the pixel structure of the plasma display rather than the texture of the moss. The question might be, is there any detail in the image that the viewer would find so interesting that he/she would want to clamber over obstacles to get a closer look? If there is, then one can make a crop of such detail and display it at full resolution as the next image in the slide show. No need to stub your toe against a piece of furniture to get a closer look.
As a matter of interest, the detail you would see in a 100% crop of a 50D image on a 50" HD plasma display (ie., no downsampling and no interpolation - the full resolution), is the same as you would see viewing a 6ftx9ft print of the full image from the same distance.