The "total image noise" is a fiction. If you don't need the high pixel count, then buy a camera with less pixels but higher quality (like the D3). If you need many pixels, then don't compare your camera to one with low pixel count.
[a href=\"index.php?act=findpost&pid=215300\"][{POST_SNAPBACK}][/a]
Gabor,
I wouldn't call it a fiction, but rather a perception. I see a distinction between pixel noise and total image noise.
When comparing noise levels on our monitor to see which camera produces lower noise at which ISO, we sometimes forget that the degree of magnification of the crops we are examining is representative of a huge print, probably a far bigger print than we would ever make from a single frame.
When comparing, for example, a 1Ds3 crop with a D3 crop, both at 100%, the 1Ds3 crop is representative of a significantly larger print than is the D3 crop. The print would be roughly 1.33x longer in each dimension.
Now, I shan't argue that one should therefore view the larger print from a distance that is 1.33x greater, because one can't dictate what viewing distance should be. People will view a print from any distance they like, and in any case, people's acuity of vision varies so much it's not meaningful to have a fixed rule about viewing distance.
But I shall argue that, when comparing noise, it's more meaningful for the viewer to examine same size prints or same size image detail in the crops on the monitor.
Let's consider the dpreview comparison of the D3, 1Ds3, D300 and 5D at ISO 3200, below.
[attachment=7963:attachment]
Each image (crop) is comprised of the same number of pixels. It's a 'pixel for pixel' comparison of noise.
The D3 image in this comparison definitely looks cleaner than the 1Ds3 image. In particular, there is less chroma noise in the D3 image. I understand these are all jpegs straight out of the camera.
But what happens if we equalise the amount of image data in each crop by interpolating the D3 image so that the Queen's face is the same size as in the 1Ds3 crop? The noise in the D3 image then becomes more obvious, but it still has the advantage of lower chroma noise because that's already been removed in-camera. That's certainly an advantage for jpeg shooters, (which I'm not.)
If we pass the 1Ds3 image through a program like Noise Ninja which can specifically address chroma noise without applying the luminance NR which tends to soften resolution, we find that the noise in the 1Ds3 image becomes roughly on a par with that in the D3 image, but the 1Ds3 image is clearly more detailed.
Some have suggested that you can do the same for the D3 image. Pass it through a noise reduction program. You can, but you can't remove chroma noise twice. The only improvement that can be made to the D3 image (according to my tests) is to apply luminance NR and soften the image even more, resulting in the 1Ds3 image being even sharper by comparison.