Nikon is making this claim: if you compare the raw data coming from individual cells off a LBCAST sensor versus a standard CMOS sensor, the base noise level is lower. You of all people, Peter, ought to know that Canon has been using the pixel-addressing ability of CMOS to apply in-camera noise correction at the photosite level for some time. The question, as I see it, is whether Nikon's pixel-addressed NR is working as well as it could. Given the assumption that LBCAST has lower noise than CMOS, the same pixel-addressed NR should result in lower noise in final images. However, Canon has their processing, Nikon has theirs, and Kodak has theirs. They are not equal, and it's a changing world, as Kodak has shown with their firmware updates.
Well, other than the use of the word "heavy," I'd say that the CMOS users already do.
Uh, no. First, NR must already be in effect. Second, we're looking at unknown samples from an unknown point in the release process. Third, the pictures that everyone keeps pointing to for noise were severely underexposed, which is a good way to see where the noise floor relates to dynamic range, but doesn't give you any information about the real noise properties of well exposed images.
Restate that as "does this mean post-camera processing NR will be required?" And the answer is unknown. The answer to that question was "yes" for the initial Kodak Pro 14n firmware, but has no made it to "no" with the latest release.
Again, you seem to be dismissing what Canon's been doing in the NR realm. The one thing that's a little troubling about Nikon switching to a CMOS-type chip is that Canon does indeed have several years shipping experience in running pixel-addressed NR correction. How fast Nikon can catch up to the quality level Canon is currently achieving is unknown, though given Kodak's remarkable strides with far fewer resources, I'd say that it shouldn't be long, if it does need to happen. (This is why the lack of user upgradeable firmware in the Nikon bodies is problematic, by the way. Kodak did it right--it's not only simple to do, but difficult to goof up.)
CCDs have better noise properties in the first place. So much so, that no one attempts to NR in the camera (for DSLRs) to my knowledge. Moreover, you can't pixel-address row transfer CCDs, so you'd have to redesign the CCD to start with or settle for DSP-based NR, which adds cost and complexity. Which brings you to some other decision points: CCDs are more costly to produce than CMOS, all other things equal. Nikon made the right decision here.
You don't work on the pixel-addressing side of things until you have the pixels working. One issue here is how fast the production line came up for the sensors and how fast Nikon got a handle on average case for those samples. And if you think Canon isn't still working on in-camera NR, you're nuts.
I'd have to partially agree with you here. Nikon's desire to use an International sporting event to get the target market sampling the camera dictated the date for launch, but not only was that a poorly chosen date in the first place, but it probably came too soon in the release processes (Nikon is now ISO 9001 certifying everything, which has to be playing havoc with the software side of things). The fact that many photographers didn't seem to understand the limitations that was imposed on picture release indicates some sort of chaos in the marketing side (if you want to restrict distribution of sample pictures, you don't hand someone a camera until they've both verbally agreed that they understand and will comply, and have signed something that says same; it appears that Nikon managed the latter, but not the former).
Well, speaking as someone who's designed the review guidelines for more than one major national publication and worked with hundreds of publications from the manufacturer side, I'll tell you that there's great debate about the wisdom of what you suggest. Indeed, there was an article that addressed this issue with Consumer Report's relationship with car manufacturers in this morning's USA Today.
I don't see how this is any different than what Canon did with the 1D. They also worked on that model for three years and the initial images they released were horrible. The images from the very first models had unacceptable banding. In response Canon made one hardware modification, added some improvements in the firmware and demanded the sensor manufacturer tighten their Q.C.
Yes, that is a grave concern of mine and everyone else who uses the 10D. All the hair in our photos look like it is melted together like a helmet. No detail, just one massive clump. That's why no one is buying the 10D, Canon sales are plummetting, and the Canon SLR forums are a ghost town. Furtunately, if I crank my 10D up to ISO 400, I can start emulating the noise/grain characteristics of the D100 at ISO 200:
--
I was just ridiculing your generalization. Obviously, a great deal depends on the lens used, the lighting conditions, the post-processing applied, the ype of sharpening used, the photographer's skills, etc, etc. This is true of any picture you see. Many users of both the 10D and the D100 have taken many a picture with plenty of detail. And many a picture has been taken with either camera with less-than-optimal detail. But rest assured that either camera can capture phenomenal levels of detail. Maybe you didn't mean to make it like such a funny generalization, but I thought that whole Canon's melting images -thing humorous.
Unfortunately the link does not work for me, so I looked at the latest 10D and D100 luminance noise comparison graph, in the 300D preview, which uses Phil Askey's new improved measurement method:
