D60 (and 10D/300D?) red blooming?

02/18/2004 12:55:19 PM · #1

Please have a look in my bloom collection (collection removed) to see images of an issue which seems to happen from time to time. There are four images, each paired with the raw converter clipped exposure view and the images themselves have the histogram below them. Notice that the red channel is completely out of whack. In one case (0813) there appears to be some channel bleeding as well (the flower "halo" which exists in the blue channel too).

Note that during shooting there was no indication clipping would occur. I hate not having per channel histograms.

Has anyone experienced anything like this or heard of it?

Edit: removed old bloom collection and replaced them with different ones. Here are the old image pairs:
Edit: removed the bloom collection again (see the bromelia posted later in the thread for a clearer example).

Message edited by author 2004-05-26 05:33:00.

02/18/2004 01:03:37 PM · #2

That's pretty wild. In the first two flower cases - is the subject actually that colour ?

in the last case the colour looks totally wrong on the frog?

Did you use flash on all of these ?

Is it the same in different converters ?

I've had occasional problems with flowers being entirely saturated in the red channel - the canon's do seem to really like blue I've noticed too.

as an example of red problems I've had too.

Don't have a good answer though. Maybe when I finish reading real world colour management :)

Message edited by author 2004-02-18 13:04:05.

02/18/2004 01:53:57 PM · #3

Originally posted by Gordon:

That's pretty wild. In the first two flower cases - is the subject actually that colour ?

The first flower's colour is definitely wrong, but the second flower's colour is pretty close to actual (relying on my faulty memory).

Originally posted by Gorden:

in the last case the colour looks totally wrong on the frog?

Yeah, that's mostly due to not adjusting the wb; the only light available was one (or two) flashlight(s) (and other unfortunate constraints too). I made the appropriate WB adjustments and added it to the collection; the blooming is improved slightly.

Originally posted by Gordon:

Did you use flash on all of these ?

0381 (no), 0743 (yes), 0813 (yes), 2175 (no).

Originally posted by Gorden:

Is it the same in different converters ?

The only other converter I've tried is canon's. The results were the same. I haven't given in to the temptation of C1 (one day perhaps!) and I recently removed breezebrowser due to disuse.

Originally posted by Gordon:

as an example of red problems I've had too.

Yeah, the red is definitely saturated there, but how was the histogram? I mean, was there any clipping? I see there doesn't seem to much, if any, bleeding. My first image (0381) has the red actually bleeding out of its "proper" place. The bleeding is particularly bad in 0813, but is evident in all images.

Fortunately, none of these images are keepers, although I like the frog picture, but that has more to do with the experience not the content (memory trigger).

thanks.

02/18/2004 08:26:46 PM · #4

.night.

Nikon AF Micro-Nikkor 60mm f2.8D

03/06/2004 06:33:10 AM · #5

I've revisited this issue. I really wanted to know how the histogram is generated by these dslrs. I'm guessing all of Canon's dslrs use the same method.

Based on a whole bunch of test images, it seems the camera's histogram is heavily weighted by the green channel. I'm assuming this is due to the bayer matrix. My conclusion is that if you're "exposing right", you're more likely to blow out the red or blue channel (if at all) than the green, so be careful if you have luminous reds (like my flowers) or blues (the sky is not).

Koriyama

Nikon D200

03/06/2004 07:38:05 AM · #6

I'm not a user, but I read this article on The Importance of White Balance on Canon EOS D30 Colour Management a few weeks ago. It may offer some perspective on the current question.

Canon EF 17-40mm f/4.0L USM

03/06/2004 09:05:07 AM · #7

Originally posted by dwoolridge:

As far as I remember there are 2x more green pixels than red or blue, so it would make sense.

Your eye is also about 2x more sensitive to green(ish) wavelengths of light than at the red or blue end (mainly because green is in the middle of the visible spectrum - there is a fall off towards infrared and ultraviolet

jimmythefish

Canon EOS-10D

03/06/2004 10:03:49 AM · #8

Could this be a problem with colour space? Converting from Adobe RGB to sRGB could potentially clip some values.

03/06/2004 11:46:49 AM · #9

Originally posted by jimmythefish:

Could this be a problem with colour space? Converting from Adobe RGB to sRGB could potentially clip some values.

If you're referring to the blooming, it's not a colour space issue (well, unless you think about the input device). The clipping, as shown in the RAW conversion histograms, is in the raw data.

03/06/2004 02:01:08 PM · #10

Originally posted by Gordon:

As far as I remember there are 2x more green pixels than red or blue, so it would make sense.

Yes, that's my understanding (incl. retina sensitivity).

So, what's the problem then?

The extra green filtered photo sensors are needed to yield {less noise,more information,better accuracy} in the green channel. Maybe Canon's histogram is just a raw dump of all the photo sensors' response: this would explain the green weighting, but not the lack of red clipping info (unless the green has more weight than 50%).

I guess I've always assumed the histogram was an accurate-as-can-be luminosity distribution (as in L*). That is, the histogram is based on calculated (interpolated) RGB values. Actually, I think that might not be an improvement (for channel clipping detection). Multi-channel histograms are really what these cameras need.

I see the real problem is the combination of luminous/bright reds/blues and a large neutral/dark green backdrop.

03/11/2004 08:35:15 AM · #11

Have a look at this image (histogram legend in details box):

It might be hard to tell at this size, but there is red channel clipping. The histogram indicates underexposure.

A subsequent +2ev shot shows a slightly-to-right-of-center exposure, but the clipping becomes painfully obvious even on the LCD (the red channel is almost completely saturated). You can see this here:

(left shows original, right is -2ev adjusted in raw converter)

A +1ev shot is also problematic with no preview indication there might be a serious problem in the red channel.

I guess everyone takes pictures of images with plenty of green in them.

03/11/2004 08:41:48 AM · #12

The solution is as you mentioned, 3 channel histograms on the camera. Otherwise you can't tell if this is happening, other than by experience.

03/11/2004 08:45:00 AM · #13

Originally posted by Gordon:

The solution is as you mentioned, 3 channel histograms on the camera. Otherwise you can't tell if this is happening, other than by experience.

I understand the Mark II has 3-channel histograms (I wonder if the lightness histogram is fixed as well...).

Do you believe it's possible to fix this with a firmware update (based on whatever information you have, however limited)?

Sigma 28-105mm f/2.8-4.0 Aspherical for Canon

03/11/2004 01:37:59 PM · #14

Based on what I know, Canon don't release firmware updates for cameras that have been 'archived' like the D60. So it may be possible with new firmware (the information is all there potentially) it seems doubtful that canon will provide it for the D60, or any of the other current cameras either.

Rooster

Canon EOS-400D Rebel XTi

03/11/2004 01:41:54 PM · #15

I was interested in seeing them but the collection was removed. How come?

GoldBerry

Canon EOS-5D Mark II

Canon EF 50mm f/1.4 USM

03/11/2004 01:43:32 PM · #16

I've used the D60 quite a bit and know people who own them and have never had this problem, sorry..can't be more of assistance.

06/25/2004 12:10:07 PM · #17

Long, very interesting discussion on how best to address the single channel 'clipping' effects, particularly in RAW conversion. Most of the commentary comes from some folks at Adobe who work on PS. In particular, the last couple of pages of the discussion pertain to these issues.

saturation masking & RAW conversion

06/25/2004 12:48:44 PM · #18

Aye, I've been following that thread lately. The stuff from Jeff Schewe (especially about the ACR 2.2 changes) is particularly enlightening. Still, I believe the channel clipping is real (in certain cases).

I did some additional reading on Bruce Lindbloom's site and also got some feedback from Jonathan Wienke on LL about my source of confusion. The luminance calculations are, of course, channel weighted by design and very, very heavily green weighted (60-80% depending on color space) and roughly between two and three times the red channel contribution (blue's contribution is practically negligible).

In other words, the in-camera histogram is based on the interpolated data (what space it used for the calculation is unknown to me) and may be correct "enough" (at most a 10% error is probably introduced if it's wrong): see my post in the LL thread here. A channel clipping indicator and/or separate channel histograms are still a must in my opinion.

Note that Jeff Schewe seems to think the channels may not be clipped at all and you can recover from them with extreme exposure compensation and a counter-balanced brightness adjustment. I'll go back and check my bromeliad shot(s) and post results.

So much to learn!