Archived posting to the Leica Users Group, 2009/12/04
[Author Prev] [Author Next] [Thread Prev] [Thread Next] [Author Index] [Topic Index] [Home] [Search]On Fri 4, Dec'09 at 1:14 PM -0500, Doug Herr wrote: > There's an interesting thread on fredmiranda.com that discusses the > tradeoffs between sensor color accuracy and low-noise high-ISO performance: It is partially true as far as I know. After all, you could could simply throw on an ND filter over your sensor, reducing signal to noise (if you don't increase exposure to compensate). CFA's reduce sensitivity similarly. However, there are many other factors which contribute to poor high-ISO performance. Personally, I'd take just about everything you read on fredmiranda about technical details and science with a grain of salt. Lots of funny logic and hearsay. Not that we don't all do that at times. But they seem to latch onto a certain idea and then try to apply it to everything, all the time. > If this is true it could help explain the DMR's relatively poor high-ISO > noise performance along with its excellent no-fuss colors. Comments and > additional technical insight are welcome. That could have something to do with it. As far as CFA characteristics go, you can see some of them in the Color Response tab of a camera's listing on dxomark.com. I'm sure that is a simplistic view of things, but it's probably better than nothing. I don't know how different the CFA's are on the M8 and the DMR, but the M8 has quite a bit of green sensitivity in the blue channel which probably can cause some problems with color rendering. (Let me see if I get this right:) It causes problems because our color systems use, for the most part, three *pure* colors, single wavelength, like RGB. However, our sensors/film don't. They see three bands of color, with varying sensitivities for different wavelengths across a given 'channel', which can be centered at different wavelengths as well from system to system. This can cause metamerism, which can make reconstructing accurate colors a guess. So for the most accurate color, we'd have very narrow bandwidth CFAs that were centered on R, G, and B. But you'd throw out a ton of sensitivity.