Magenta Color Cast with Heavy ND Filtering on 5D MkII

Hey all,

I'm a bit hesitant to post this - I don't want to start another "5D MkII" debate...

I'd like to share a discovery I made, and want to know whether anyone with a 5D MkII or 1Ds MkIII has seen this before.

Here's the scoop: when I use a Singh-Ray 4-stop ND filter with a Singh-Ray 3-stop GND filter, I get an extremely heavy magenta cast in the double-filtered section of the image. We're talking a -75 to -90 point tint adjustment in ACR to correct the magenta.

Both the filters appear to be very neutral - I cannot detect any magenta by eye even when stacking them, and used on their own, there is no evident magenta cast in my test images. So the I am not blaming the filters.

I've done extensive tests shooting white snow under even overcast conditions, as well as shooting a Color Checker chart indoors (natural light only). The tests were done using different filter combinations, and my results are consistent and repeatable. With 7 stops of ND filtering, the 5D MkII shows a strong magenta cast, but with 6-stops of ND filtering (2-stop GND + 4-stop ND), the magenta cast is much less evident, requiring a tint adjustment of "only" -25. Using my polarizer has no obivous effect either way - so using a 2-stop GND + 3-stop GND + polarizer, and I don't get much of a cast - only about -10 points of adjustment required.

Here's the kicker though: when I use exactly the same setup on my EOS 20D, with the 3-stop GND plus 4-stop ND, there's very little magenta , with only a -12 point tint adjustment. Considering the 7-stops of filtering, this amount is trivial.

So it appears that the problem is with the 5D MkII, not the filters.

I've been exchanging emails with a well-known photographer who uses these Singh-Ray filters, and who frequently writes posts on the Signh-Ray blog site. He's also investigating this issue, as he's noticed the same magenta cast when using his 5-stop ND filter (both with and without a GND) on his Canon 1Ds MkIII. As you know, these cameras share a similar sensor. He stated that he has not seen this problem using his Vari-ND, only with his regular 5-stop ND. My wild conjecture: the Vari-ND uses the same type of materials/construction as a polarizer - I've not seen the magenta cast change in any way when I use a polarizer, so maybe this could explain it?

Has anyone seen something this heavy magenta cast using heavy ND filtering on their Canon cameras?

Is this something to do with the sensor? Sensor filters? Firmware? Any guesses?

Did you manually set your white balance or use AWB ? Perhaps with Grad filters auto white balance interprets wb incorrectly leaving a cast. Just a thought.

Neither.
I'm checking white balance in ACR, using the white balance eye-dropper tool on either the white snow or a color checker chart. I'm then comparing these values with the "no filter" shots. There is a 75 to 90 point "tint" difference between the 7-stop ND filtered image and the non-filtered one on the 5D MkII. There is no significant tint difference with the same setup on my 20D, regardless of the filters I use.

I would check the WB and as Christian said because AWB does not do a good job with a lot of ND. I have the same issue when using a variable.

Guys - my problem is not about AWB. I am not using AWB, nor do I care about it. My problem is about a very heavy magenta tint when using the 3-stop GND + 4-stop ND on the 5D MkII. When I compare the same subject with the same filters and lens taken on the 5D MkII and the 20D, there is no magenta tint to the images taken on on the 20D, but there a crazy cast on the 5D MkII. The image on the 5D MkII requires a "Tint" adjustment in ACR of about -85 - that's just crazy.

Now, I can actually deal with this tint as such. But the problem is when I use the 3-stop GND to just darken the sky: my sky is magenta, but the rest of the image is "normal". So if I use global white balance (tint) adjustments in ACR, I either have the choice of going for a magenta sky and a normal foreground, or a crazy green-tinted foreground with a normal sky.

I know that I could double-convert the image with different tint settings, and merge them in PS. But that would defeat the purpose for which I purchased filters: I want to do the work in the field with my camera and filters, not in PS...

So... My question is not how to color correct or white balance the images. My question is whether anyone has encountered this themselves, or has any insight as to the possible cause.

Rob, does this cast appear when you look at the back LCD when reviewing the shot or using live view? If it's something optical being triggered by the stacked ND's, it should be visible on the LCD. If you are getting a magenta cast in the sky even just with a single 3-stop GND, that normally would imply to me that the filter is not actually neutral. But since you can use the same filter on your 20D without seeing the sky go magenta, the camera has to be a player in this as well.

I'm wondering about some kind of difference in the layers over the sensor or the microlenses or something like that. The 20D is by now a pretty old sensor design, and a fairly simplistic one too compared to what is in your 5D Mk II. I wonder if the latter could have "broken" the filters due to some optical interaction with the filtered wavelengths of light that are coming through...

Royce Howland wrote:Rob, does this cast appear when you look at the back LCD when reviewing the shot or using live view?

Interesting question... I will need to test the live view tomorrow in natural light and will report back. But reviewing the image on the LCD shows a very strong magenta cast when either AWB or "Sunny" WB was used and the two filters are stacked.

Royce Howland wrote:If it's something optical being triggered by the stacked ND's, it should be visible on the LCD. If you are getting a magenta cast in the sky even just with a single 3-stop GND, that normally would imply to me that the filter is not actually neutral. But since you can use the same filter on your 20D without seeing the sky go magenta, the camera has to be a player in this as well.

On the 5D, there is no color cast using only the 3-stop, and only a very "mild" cast with only the 4-stop. With a 2-stop plus the 4-stop, there is a "medium-light" cast (tint at about -20). The problem only really occurs at 7-stops of ND, where the tint correction is somewhere between -75 and -90.

I would also think that I should be able to detect such a strong cast if it were optically caused by the filters. But of course, my tests on the 20D discounts any cast being optically generated. This must happen on the sensor itself, which means that I should be able to see it in live view. More tomorrow. Thanks for metioning this.

Royce Howland wrote:I'm wondering about some kind of difference in the layers over the sensor or the microlenses or something like that. The 20D is by now a pretty old sensor design, and a fairly simplistic one too compared to what is in your 5D Mk II. I wonder if the latter could have "broken" the filters due to some optical interaction with the filtered wavelengths of light that are coming through...

I am inclined to believe that it is something along those lines. It'd also explain why this happens on the 1DsMkIII for that other photogapher, but not on my 20D.

If it is either a sensor or firmware processing problem, what is the process to notify Canon of this?

RServranckx wrote:If it is either a sensor or firmware processing problem, what is the process to notify Canon of this?

All humor aside, if it is the camera, the procedure is probably "bend over and kiss your filters good-bye." If it's an optical interaction, there's no way Canon doesn't already know about it and there won't be anything you can do about it except to correct in post-processing as previously discussed -- including dual conversions when GND's are involved. There may be other filters that work better of course, if some non-neutrality in the filter glass is contributing. But the bottom line is if it's the sensor or imaging pipeline, then that's the way it is. More HDR work in your future.

We'll see if any other 5D Mk II or 1Ds Mk III users comment...

Royce Howland wrote:All humor aside, if it is the camera, the procedure is probably "bend over and kiss your filters good-bye." If it's an optical interaction, there's no way Canon doesn't already know about it and there won't be anything you can do about it except to correct in post-processing as previously discussed

Yeah, I realize that you're probably right. And as there are very few photogs who use 7 stops or more of ND filerting (excluding the Vari-ND, which does not seem to cause the problem), it's doubtful that Canon would care. Darn...

My only hope is that this "other" photographer, who's well-know and well-published, may have a bit of clout, although probably not a whole lot with Canon.

I suspect I know who you're talking about. But regardless, I think it's safe to say there is absolutely zero chance of getting Canon to do anything about this, if it is a camera-filter interaction. It would be more likely to have some clout with Singh-Ray and have them go back to the drawing board on their filter design to see what they could do to minimize the interaction.

I use a B+W Ten-stop ND filter all the time on all of my cameras, (5Dmk1, 1dsMk2, 400D, etc) and I have ALWAYS gotten a magenta color cast or a purple-ish hue in what should normally be a neutral blue sky... I never figured it out, always figured that's just 'the way it is' - but to correct it, I've turned the tint way down in camera raw a few times to compensate...

I can't say why you see more magenta when using your 5DII as opposed to your 20D. My guess is that because the 5D records more pixels and more information, the effect just becomes more pronounced.

From experience, I have noticed an extreme magenta cast when using a Singh-Ray 3 stop hard grad and Singh-Ray 3 stop solid ND filter in combination on a 20D. The camera was set to auto white balance, and it was never tested if using a preset white balance would fix the issue, even though I suspected that had quite a bit to do with it. That camera no longer works, so the test cannot be done. Alone, both Singh-Ray filters seemed to work fine on the 20D without a color cast, and the filters themselves appear to be neutral.

I don't know the exact root of the problem (I still suspect auto white balance) but I would not jump to the conclusion that it is a flaw specific to the 5D Mark II.

Royce Howland wrote:Rob, does this cast appear when you look at the back LCD when reviewing the shot or using live view?

I tried this just now. It is difficult to be sure, as the live view image on the LCD is rather dim with all of this filtering - I've boosted the LCD brightness to compensate, but this will obviously skew the results. But yes, there is a cast visible, although it does not strike me as being as pronounced as when reviewing the brighter/properly exposed image on the LCD.

California4Life wrote:I use a B+W Ten-stop ND filter all the time on all of my cameras, (5Dmk1, 1dsMk2, 400D, etc) and I have ALWAYS gotten a magenta color cast or a purple-ish hue in what should normally be a neutral blue sky... I never figured it out, always figured that's just 'the way it is' - but to correct it, I've turned the tint way down in camera raw a few times to compensate...

Thanks for the info, Mac... I'd heard of this strong magenta cast when using this specific filter - I had always believed that it was due to the filter not being neutral, but given my results, it may well be a similar issue to what I am seeing. I am very puzzled by the fact that my 20D does not have this magenta issue in my tests. For what it's worth: with slightly different lighting conditions today, my snow is now more purplish/lilac than magenta... So it's likely that a combo of the light temperature (blue/yellow) along with the magenta cast cause some funky colors in your blue skies... You can see one of these "heavy magenta sky" images here: http://www.naturescapes.net/portfolios/ ... pos=-24163 Notice that the 3-stop GND was only on the sky (well, and on a tiny bit of snow - got to place that GND filter more carefully!). Now, in this image, I like the magenta.

Kari Post wrote:From experience, I have noticed an extreme magenta cast when using a Singh-Ray 3 stop hard grad and Singh-Ray 3 stop solid ND filter in combination on a 20D. The camera was set to auto white balance, and it was never tested if using a preset white balance would fix the issue, even though I suspected that had quite a bit to do with it. That camera no longer works, so the test cannot be done. Alone, both Singh-Ray filters seemed to work fine on the 20D without a color cast, and the filters themselves appear to be neutral. I don't know the exact root of the problem (I still suspect auto white balance) but I would not jump to the conclusion that it is a flaw specific to the 5D Mark II.

It's interesting that I do not see any significant cast on my 20D, so our results are very diffirent... Very puzzling. I am not saying that the problem is specific to the 5D MkII - it certainly looks as though it's also present on the 1Ds MkIII as per another photog. It's difficult to guess whether Mac's problems is due to the B+W 10-stop filter being off, or an issue with the cameras; at 10 stops of filtering, it's tough to gauge whetehr the filter is really neutral - but the odds are that this may be a Canon sensor or firmware image processing issue.

As for AWB: this could be the case when reviewing the image on the LCD. But I am also checking the tint of the snow using the WB eye dropper in Adobe Camera Raw, using the snow as a reference point. So AWB does not come into play for these measurements - I can clearly see that there is a 75 to 90 point "tint" difference (towards magenta) with the 7-stop filtered image compared to the unfiltered image.

Hum... I'll probably alert Singh-Ray of these results. I don't know whether they can do anything, but it'd be great if they could conduct their own tests.

Thanks all...

I think you are assuming that the sensor (and bayer filter) has a linear response to different frequencies of light. ie: as you reduce all the frequencies to a lower level evenly, the sensor response to the light is the same.

What you are seeing I've seen with 4 and 5 stop ND Singh-Ray filters on various Nikon cameras.

Anthony Medici wrote:I think you are assuming that the sensor (and bayer filter) has a linear response to different frequencies of light. ie: as you reduce all the frequencies to a lower level evenly, the sensor response to the light is the same.
What you are seeing I've seen with 4 and 5 stop ND Singh-Ray filters on various Nikon cameras.

Hey Tony,

Yes, that was certainly an assumption I had been making... Apparently an incorrect one! I was just so surprised that the problem is not evident on my 20D but it's there on my 5D MkII.

I found the following info on the Singh-Ray FAQ page:

Q13) I'm getting a color cast with my Neutral Density filter. What's the problem?

A: When using a Neutral Density filter (graduated or solid) you are generally causing your camera to make a longer exposure than normal, sometimes much longer. The software in some digital cameras uses response curves that adjust the RGB sensors sensitivity based on the exposure time, which may introduce more red into the image. Sort of a digital form of reciprocity failure. In other instances, blocking visible light and a longer exposure allows infrared light to "build up" on the red sensors leading to a red or purple-ish cast. If your filter looks neutral but you see the cast in your images, try taking some test shots using the filter with a different brand of camera, or with a film camera. If you still see the cast, and you're within the warranty period, contact Singh-Ray to have your filter evaluated and possibly repaired/replaced

I sure this is not related to exposure time (or at least not completely related), as I am getting the same color cast at 1/5s as I am at 30s with the filters, but no cast without the 4-stop ND regardless of exposure time. But as the problem is only on my 5D and not my 20D, I clearly cannot blame the filters.

You could try adding an IR blocking filter to see if it is an issue with the IR frequencies. Other than that, all you can do is change the WB in camera or in post processing to compensate.

It may be that the transmissivity of the filters is not as "neutral" as they would have you believe. Since camera spectral response curves are different from those of human vision (leading to all sorts of metamerism issues), this might account for the tint being observed. Tony's suggestion of an IR cut filter thus sounds reasonable; one might also try various other color filters in combination to restore a balanced spectrum to the light as the camera sees it.

It's possible that the composition of the CFA color dyes might have changed a bit between the 20D and the 5D so that the filter over the "green" channel in the newer formulation does a slightly better job of blocking the (invisible to the eye) IR tail of the spectrum than do the red and blue ones. In this case we'd also have to speculate that some NDs let more IR through than others. When the visible light is so heftily attenuated with the ND, the residual IR components, could come into play (even with the IR-block filter on the AA filter assembly) more - thus the R and B channels could get a boost solely because their filters don't block the IR quite as well as the G. The suggestion to add more IR blocking would tell us if this is on track. All pretty speculative but, from a lifelong spectroscopist, it's the best I could come up with to fit to the observations

Toney, Emil and AJ - thanks for the very interesting information... I accept this as being very plausible (as the Mythbuster guys would say!), and it does make sense to me. I don't think adding an IR filter to the mix is for me (already using ND + GND + pol), but it would be very interesting to find out whether this affects the cast...

I wonder why the Singh-Ray Vari-ND apparently does not cause this problem - I suppose that it blocks certain wavelengths, thereby avoiding the problem... Can someone check to confirm that the Vari-ND does not cause this magenta cast?

The problem with the Vari-ND is it vignettes as you pass a certain level of ND. The effect is different depending upon focal length and, of course, depending on the lens, you could end up seeing the Vari-ND. I agree that adding filters rarely works and I've been working out says of not using filters at all.

For Grads, simply using an HDR technique would remove the need to use one. For the ND, the Nikon cameras allow in camera multiple exposures where you can balance the gains. With enough images blended, the effective shutter speed becomes very slow. This could be done out of camera at the cost of post processing images to get the same effect. That leaves the polarizers which I'm only using to remove reflections. If I want to polarize a sky, I do that in post production.