E-M5 sensor nonlinearities

Started Jan 5, 2013 | Discussions thread
Anders W Forum Pro • Posts: 21,466
Re: E-M5 sensor nonlinearities

gollywop wrote:

Detail Man wrote:

gollywop wrote:

Detail Man wrote:

Anders W wrote:

gollywop wrote:

Thanks, Anders, for the citations. I recall the thread fairly well, but didn't have a quick source to link it.

I guess there are alternative explanations for the behavior. The sensor could simply be linear, or it could be nonlinear at full sensor saturation but engineered to have raw saturation that stops short of this nonlinearity, or the camera could be nonlinearly encoded to counteract any sensor linearity. These possibilities would seem to be indistinguishable from what we can "see on the outside."

However, it really doesn't make any difference. As long as ETTR doesn't result in meaningful (any) color shifts, I'm satisfied.

Yes, all we know based on your test along with my findings (as reported in the post I linked to) is that the ADU response is linear, which need not mean that the analog sensor response is linear. However, we can and do know a bit more than those results alone tell us.

It seems to me that if the E-M5's color-filtered photosites were non-linear within the range of illumination used in his tests, and their outputs were as a result linearized prior to being combined in the RAW-level R, G, and B channel image-data, then (perhaps) gollywop might have seen some evidence of that ?

No I didn't; but I can't rule that out.

Or would he have seen such evidence when using the technique of only examining the RGB values of the particular colors that clipped (as opposed to examining the RGB values of all colors, as was done in the case of the white square clipping) ?

Excepts from gollywop's original post (underlining added for emphasis):

I set up my CCPP and took a series of shots (M mode, f/5.6, ISO 200) with shutter speeds that went in 1/3 EV steps from severely underexposed to a complete blow-out of all 24 squares.

The white square, D1, was, of course, the first to blow. I then examined several images starting with the one that was one exposure level less, in which the white square was 1/6 EV below clipping. Toggling through them, I could see no discernible differences in any of the colors, but, just to be sure, I used the eye dropper to measure and compare the RGB trio for about a dozen individual colors. They were all the same within a point or two.

However, it appears that the RGB values of all of the other colors may not have been examined (as was the case when the white square clipped):

I just checked all the colored squares in the case where the white square clipped, and they are all the same as you go down in exposure. In the other cases, where some of the colored squares had clipped, of course the clipped colors would no longer be the same, but the unclipped ones appear to be, at least until most of the image is blasted.

I was kind of wondering out-loud whether linearization (of the RGB color-filtered photo-site outputs) could be detected by such tests - but upon some futher reflection on my part, it seem (to me) that such linearization processes could not be detected by the particular tests that you performed.

Yeah, I don't see how they could.

It seems like linearization could only be detected by testing SNR - and the resulting effects upon color would relate to effects upon the coloration of image-noise. Those effects would only be relevant to the extent that they might influence the coloration of (combined) recorded signal and image-noise.

Right. But couldn't you use RawDigger to check the s/n in individual color channels? Didn't Anders do something like this?

Yes, that's what I did.

-- hide signature --


 Anders W's gear list:Anders W's gear list
Panasonic Lumix DMC-G1 Olympus OM-D E-M5 Olympus E-M1 Panasonic Lumix G Vario 14-45mm F3.5-5.6 ASPH OIS Panasonic Lumix G Vario 7-14mm F4 ASPH +28 more
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow