In your analysis, D2X has lower luminance noise than D2H & 1D mark II , really ? It's hard to believe that 5.5-micron Sony's CMOS sensor could outperform both 9.6-micron Nikon's JFET LBCAST sensor and 8.2-micron Canon's CMOS sensor in terms of noise !! R you kidding me ? BTW, according to Phil's test, lum. noise of 1D II is less than D2H's.
Http://www.dpreview.com/reviews/canoneos1dmkii/page18.asp I dont know whose results are correct...
Actually, I am not surprised. From DP's review, "Visible difference between the Canon EOS-1D Mark II and Nikon D2H are only apparent from ISO 800 upwards, and even then it's very subtle and unlikely to be enough of a difference to be noticeable in a normal shot." On the other hand, the noise characteristics of the D2X are superior to those of the D2H, and so... I am not surprised at all. Please note that the noise characteristics are not just defined by the individual sensor photosite size and spacing, but also by the processing electronics, and both are state-of-the-art on the D2X. Noise studies are always complex, though, and possibly different techniques should be used in parallel to get a good understanding of what is happening not just with the D2H but with all other cameras. Thanks for the great review, Dave! Best regards,.
In the high resolution test shots of the prototype D2X there seems to be a lot of colour fringing, especially in the shots using very short focal lengths and with high contrast. I guess this camera shows up any lens' tendency for chromatic aberation. The fringing could be cleaned up afterwards but do you forsee the camera's firmware helping with this issue at all. Best regards, Steve..
Grape - Well, as I've pointed out many times, the simple magnitude of luminance noise tells only a small fraction of the full story with image noise. There are a couple of things to consider: 1) Noise from the sensor accounts for only part of the total "noise budget" in a camera. The amplifiers and readout circuitry, and A/D converter also contribute noise. (Some of what Jose was talking about in his subsequent post.) 2) ALL digital cameras do some level of noise-suppression processing, to mitigate the effects of sensor & system noise. This makes a huge impact on the final output. The trick is how much noise suppression they achieve, and at what cost to subject detail in areas of subtle contrast.
When the sensor data falls below that threshold value for local contrast, the noise-suppression algorithm becomes much more aggressive, flattening out the inter-pixel luminance differences. (It's basically saying "OK, if I've got less than "x" amount of contrast between pixels, then the small variations I'm seeing must be the result of noise, so I'll just eliminate them.) The problem with this approach of course, is that when you flatten out inter-pixel differences, you're also potentially flattening out subject detail as well. The trick is to pick the threshold value for contrast and the area over which the flattening occurs to eliminate as much noise as possible, without killing off important subject detail. FWIW, the D2x generally does a pretty good job with this, although I haven't yet shot the test images where this issue is most obvious. (The "Sunlit" Portrait shot of Marti - Blonde/brown hair is excellent for seeing where the anti-noise algorithms are losing subject detail.
With some cameras, the bricks that are in shadow look like they were finger-painted) To see an example of a camera that gets pretty heavy-handed with noise suppression, check out the Far-Field or Indoor Portrait shots from the Konica Minolta Maxxum 7D at high ISO. The 7D's noise magnitude stays almost constant from ISO 100 to 1600, but the loss of subject detail at ISO 1600 is pretty horrendous. This by the way also gives the lie to the practice of looking at image noise in flat tints like sky shots and MacBeth targets. Unless you shoot a lot of sky areas (which some photographers obviously do), or large expanses of flat walls, these subjects have almost no relevance to what you're going to see in actual photos you might take. Picking apart RGB channel noise in swatches from a MacBeth Color Checker may be intellectually entertaining, but it isn't going to tell you anything about how your model's hair is going to look the next time you shoot a portrait in the studio.
Or brick or stucco patterns on your next architectural subject. Etc, etc.) Another thing cameras do to knock down image noise is to cut the color saturation. Too much to go into here, but if you reduce the saturation in the initial processing of images from an RGB sensor, you'll generally reduce the image noise as well. This was something that the D2H did, that left it's images looking distinctly second-best when compared to those from the original EOS-1D, despite the D2H's having lower absolute noise levels. A final issue that contributes to perceived (vs measured) noise levels is the frequency content of the noise, the "grain structure," if you will.
I've made these points here and there in my reviews in the past, I should probably do a whole article just on image noise, given how misunderstood it is. The fact is, simple plots of luminance noise magnitude vs ISO that Phil and I do are almost totally worthless for telling much about what a camera's images look like. I've started doing them for SLRs that I review because so many people kept asking for them, but I do so with a lot of misgiving, as I think they do more to mislead people than to inform them. So about all we're left with as being valid are very subjective measures of noise, just personal judgement calls of how images from various cameras look to various people. FWIW, images from the D2x look very good up to about ISO 800, then start to get rougher from there.
But even there, I need to caution people about just looking at images on-screen and basing their purchases on that. I've found that printed output leads to very different impressions of image noise, and that the impressions there also have a lot to do with the printing process used and the size that you're outputting at. This is why we've now started routinely printing multiple images from the cameras we test on the Canon i9900 high-end inkjet printer in our studio. Taking images from the D2x as a case in point, ISO 3200 images from it look pretty awful printed at 13x19, but surprisingly OK at 8x10. At 5x7, they're better yet.
The bottom line is that noise is a very complex issue, and trying to compare single numbers as a way of evaluating the relative merits between cameras is almost totally worthless. - And looking at RGB channel noise on MacBeth targets is only (very) slightly better...
Stephen - No question about it, the smaller the pixels get, the more critical the lens quality becomes. I realized we didn't note the lenses used in those shots, will try to get back to include that info where possible. Other than the studio shots and the shot of the pine tree (which used the 100mm f/2.8 Micro Nikkor), the shots used a mixture of the 12-24mm f/4 DX and 17-55mm f/2.8 DX lenses, both apparently fine lenses. - BUt you're right, high-resolution d-SLRs definitely stress the optics a lot more than film ever did...
Thanks for the excellent comments on noise, Dave. It is certainly a very complex issue that few people really understand to the level required to have very meaningful discussions, myself included. Regarding the additional demands that an "optical sensing system" like the one that D2X incorporates makes on lenses, this brings up the subject of how relevant may a telecentric lens design be for use with a camera body like this. I fully agree with you that the D2X imposes tougher demands on lenses... and it will be interesting to revisit the "lens quality" concepts in the context of their use with the D2X. Thanks again for your excellent review.
Thank you very much Dave. I was wondering because I've seen many D2X's test but I can't feel it's cleanness especially in the range of ISO 800 - 3200. BTW, I wouldn't be much surprised if you set noise reduction to High, didn't you ? See "ISO Composite JPG" at the bottom of Ron Reznick's D2X tests.
Http://www.digital-images.net/Gallery/Nikon_D2x/nikon_d2x.html Personally, I don't much appreciate Canon's style that is low noise but too soft. Yes. I know the reasons...
Seems to me that some objective image quality and noise evaluation should be done under tungsten illumination with tungsten (or equivalent manual)camera settings. The needed blue channel amplifier gain causes noise in larger amounts over daylight configurations. Many folks shoot tungsten professionally and for fun. tony.