For certain images, the green split for Sony files really is very large, and the value of 10,000 was required to prevent artifacts.
I checked out a bunch of A700 raw files. Those without BayerGreenSplit are the exceptions, most of them - the lossless ones as well - get converted in DNG with 10000.
1. I measured how well the greens are balanced in images from six different copies of A700, multiple shots per camera. The differences are from 0.01 EV to 0.16 EV in the very shadows, and 0.00 EV to 0.03 EV overall. Notwithstanding the actual differences, the BayerGreenSplit is always 10000.
In other words, the BayerGreenSplit value has nothing to do with the actual image content and imbalance between the green channels.
2. I measured images from other cameras as well. I found for example a Canon 5DMkII image with imbalance between the greens 0.25 EV; the BayerGreenSplit is 0.
3. I modified the BayerGreenSplit value of an A700 image from 10000 to 0.Converted by ACR, Black 0, NR 0, there IS difference, the version with value 0 does exhibit some artifacts. However, the difference is almost exclusively in the red and blue channels, it is negligable in the green channel between the two versions.
4. The specification of the BayerGreenSplit tag is so vague as to be useless:
This tag specifies, in arbitrary units, how closely the values of the green pixels in the blue/green rows track the values of the green pixels in the red/green rows
a raw converter would have to have at least some idea of the meaning of the values. Beside, any raw processor can evaluate the balance between the green channels, there is no need to be told about that. (If "track" has a different meaning in the specification, then what is that?)
I have the feeling, that the DNG converter and ACR are using this tag in a proprietory, undocumented way.
Well, in my experience, the green channel split isn't just about average difference, its also about "inter-channel noise".
But I'd like to add my voice to Gabor's in asking for a better specification of the BayGreenSplit. While I understand that in practice the value has probably been set by practical "when do we get artifacts" testing, I don't think that's good enough for an open standard. The issue is that while the practical test approach may work for Adobe, using an Adobe demosaicing algorythm, it doesn't help anyone using another algorythm.
So, my request is that Adobe apply some of its engineering resources to developing a workable metric - I'd assume some kind of statistical measure - that ties this value down.
the green channel split isn't just about average difference, its also about "inter-channel noise"
I don't know the meaning of "inter-channel noise", but when I analyze the noise characteristics, I separate the green channels and see how they behave. Better cameras, like the Nikon D3, Canon 1DMkIII do not exhibit relevant differences, except faulty copies. I have black frame shots from the A700 as well, and I don't see any noteworthy difference at ISO 100, but that image still gets BayerGreenSplit 10000.
Btw, if the color filters and microlenses are identical over the two greens, the average of the two greens over a larger area of any image has to be close to identical (I measure this on areas of 10,000 to several million pixels), if the A/D is well balanced. Images from a few copies of the Canon 5DMkII exhibit excessive artifacts; I found, that the green channels behaved very differently. It turned out, that the board was faulty, the reading and A/D was imbalanced (the 5D2 reads four columns at the same time).
It's the local deviation, not the average deviation, in other words. So large areas aren't that important. Alltough, of course, large areas are more visible.
It's the local deviation
The differences between the two green pixels within a CFA on a uniform area, where they are supposed to be equal (within the expectable noise) are not worse with the A700, sometimes even better (at low ISO) than some other cameras, for example the Nikon D300.
Same here. Sony might have some DSLRs nowadays and sure they might be cheaper and I'll even agree they are a good start, but Sony has alot more work to do before they are up to par with Nikon, Canon, or even Pentax. I much prefer the layout of buttons etc Nikon uses, not so much how Sony handles it. I've used a couple Sony's just messing around with them in stores and like I said they still need a lot of work. However, at the end of the day its the photographer that makes the photo, not the camera, although the camera either detracts or adds to the creative process.Shirt |Jeans |Kaffee |Dakine |Armour |Sangenic |Tatonka |Nintendo |Dakine |Golana |Salewa |BHs |Kindle |Redolz |Dartbox |Teleskop |Geschirr |Kissen |Camping |Rucksack |Pinolino |Hunde |Dachbox |SPRÜHER |Shop GPS |