Normal colors test: Tangsinuo QB21 & ICF ("hot mirror") filters on Sony a7 full spectrum camera

[LarsHP]

 

I have bought QB21 and ICF filters from Tangsinuo on eBay for normal color imaging with my Sony a7 full spectrum converted camera with EL-Nikkor 80mm f/5.6 (old metal version) and found them, particularly some of the QB21 filters, to be too thin, which results in reds that are too light and signs of IR leak. After receiving yet another set of 52mm and 58mm filter that were listed as 2mm QB21, I thought they looked just as thin as before, but this time I did more than just test images of ordinary subjects.

 

First of all, I took out the filters of the frames, and indeed, the 58mm was not 2mm, but 1.5mm. To my surprise, the 52mm filter was actually 2mm. I complained about the 1.5mm thickness to Tangsinuo, and made an agreement that I could order other filters, and then he could ship a 2mm thick glass with them. I am now waiting for these.

 

Today, I took my X-Rite ColorChecker out in the spring sunlight here in Tromsø and did a test series with no filter, 2mm filter, 3mm (listed as 1.5mm ICF and 1.5mm QB21) and 4mm (listed as 2mm ICF and 2mm QB21). Jason said in a eBay message to me, that ICF and QB21 are the same. However, the official transmission graphs vary a bit in the 400 to 500nm range, while they appear to have roughly the same, but not quite the same, cut in both ends of the spectrum. See the two first graphs below.

 

Below are the test images ordered like this:

No filter, white balance in camera set to daylight (sun symbol)

No filter, white balance set in Adobe Camera Raw

2mm ICF, white balance in camera set to daylight (sun symbol)

2mm ICF, white balance set in Adobe Camera Raw

3mm (2 x 1.5mm QB21), white balance in camera set to daylight (sun symbol)

3mm (2 x 1.5mm QB21), white balance set in Adobe Camera Raw

4mm (2mm ICF + 2mm QB21), white balance in camera set to daylight (sun symbol)

4mm (2mm ICF + 2mm QB21), white balance set in Adobe Camera Raw

 

I realize they are overexposed, but for a relative comparison the test should be valid.

 

Forgetting about the unfiltered, which is just included for reference, it appears obvious to me that even the "correct" 2mm thickness is too red and leaks a bit of IR as seen in the completely neutral looking black cloth.

The 3mm stack looks much better, and is the best to my eyes.

The 4mm stack looks too blue in the camera set WB, but is fully usable.

 

Note that camera manufacturers have different cut in the red end. Sony cuts the most red, so on a Nikon or Canon the 2mm thickness may fit just fine. See the last graph (which is from Kolari Vision).

 

 

Tangsinuo QB21x2.0mm Chart

 

 

Tangsinuo ICFx2.0mm Chart

 

No filter, white balance in camera set to daylight (sun symbol).

 

 

No filter, white balance set in Adobe Camera Raw

 

 

2mm ICF, white balance in camera set to daylight (sun symbol)

 

 

2mm ICF, white balance set in Adobe Camera Raw

 

 

3mm (2 x 1.5mm QB21), white balance in camera set to daylight (sun symbol)

 

 

3mm (2 x 1.5mm QB21), white balance set in Adobe Camera Raw

 

 

4mm (2mm ICF + 2mm QB21), white balance in camera set to daylight (sun symbol)

 

 

4mm (2mm ICF + 2mm QB21), white balance set in Adobe Camera Raw

 

 

 

 

[Andy Perrin]

When you white balance in ACR, did you use the third gray on the color checker from the left?

[LarsHP]

1 minute ago, Andy Perrin said:

When you white balance in ACR, did you use the middle gray on the color checker?

Since I overexposed, I used the gray patch next to the black. However, all the gray patches are equally neutral after WB, so it doesn't matter, I think.

[Andy Perrin]

It matters that it not be too dark (because of color noise) or too light (because of saturation of the channels). 

[LarsHP]

12 minutes ago, Andy Perrin said:

It matters that it not be too dark (because of color noise) or too light (because of saturation of the channels). 

In that case, I guess I selected the best patch. 

[Andy Perrin]

Yeah. 
—

 

I think the second pair are the closest to not changing after whitebalance (but be aware I’m a bit colorblind). None of them look like a perfect fit for restoration of the original camera colors.

 

In past experiments by Andrea, it was found that Schott BG38 2mm was a good choice. 

[dabateman]

4,5, 6 maybe look good.

 

Every camera is a little different for best correction. My Olympus E510 was great with B&W 486 filter. But not the m43rds cameras. 

For my Em1mk1 I think 4mm BG38 I had was best. For my Em5mk2 it was something else, maybe BG40 at 2mm thickness. I can't remember off the top of my head. 

The Sigma SD14 was best with the SD15 clip in filter.

[photoni]

@LarsHP I wonder how it is possible to make comparisons and make judgments on soft, overexposed files... and without a color profile.
the last one with this correction curve looks good but still desaturated in the light boxes.

.

[LarsHP]

46 minutes ago, photoni said:

@LarsHP I wonder how it is possible to make comparisons and make judgments on soft, overexposed files... and without a color profile.
the last one with this correction curve looks good but still desaturated in the light boxes.

.

Since I used "daylight" (sun symbol) in-camera preset white balance in direct sunlight, I can get a quite good pointer regarding how close the filter is to what the camera originally had: The gray patches should be close to neutral. In addition, the black cloth reveals IR leak easily. These target subjects alone will tell me whether the in-camera preset white balance is off or not. 

 

However, I agree that the color patches are too flat and de-saturated compared to how it should look, so these aren't useful, but don't need to be either, since I have the in-camera preset white balance as a reference point, when using the corresponding light source. (Flash and artificial light would probably not work as reliably.)

 

So, ideally, the filtering should be spot on, when using direct sunlight as light source and daylight in-camera preset. If WB is only slightly off, then the filter is close to the native hot mirror filter, and a post process WB should be sufficient. That said, I plan creating a custom camera color profile in order to get the best result. 

 

The reason posting my findings here was mostly pointing out that the Tangsinuo ICF and QB21 2mm (BG38 equivalent) filters are too thin for Sony cameras. 

[LarsHP]

Okay, for completeness’s sake, I have re-edited the raw-files by reducing exposure to a more correct level and posted here.

 

For each filter, the first image is with in-camera white balance (daylight / sun symbol), and the second with white balance set in Adobe Camera Raw.

 

The first two are with the 2mm ICF filter, then the two with 3mm (2 x 1.5mm) QB21, and finally the two with 4mm (ICF + QB21).

 

My conclusion is the same as before:

2mm ICF filter is too thin for a Sony camera. The black cloth shows obvious signs of infrared leakage.

3mm QB21 appears to be closest to what the camera originally had. The red and blue patches look about right.

4mm (ICF + QB21) is a bit too blue, with the red patch a bit dull, but with a custom camera color profile, I expect this to be workable.

 

 

2mm ICF filter:

 

 

3mm QB21:

 

 

4mm (ICF + QB21) :

 

[photoni]

maybe my monitor is not perfectly calibrated, but yours has problems, the histogram says there are no blacks
the darkest black square on the scale is approximately 67-62-62

.

.

today out of curiosity I photographed the filter removed from my Sony A7 and some similar filters with flash light.
I wrote the thicknesses, the original one is the thinnest

.

.

[LarsHP]

13 hours ago, photoni said:

maybe my monitor is not perfectly calibrated, but yours has problems, the histogram says there are no blacks
the darkest black square on the scale is approximately 67-62-62

.

.

today out of curiosity I photographed the filter removed from my Sony A7 and some similar filters with flash light.
I wrote the thicknesses, the original one is the thinnest

.

.

 

There is absolutely nothing wrong with my monitor. It's an Eizo ColorEdge display, hardware calibrated with X-Rite i1 Display Pro. In other words, it's as close to perfect as it gets. However, the darkest grayscale patch isn't black, because it isn't and shouldn't. 

 

Maybe I could have enhanced contrast further, but the images I have posted have more contrast and saturation than how the ColorChecker looks in real life.

 

The lens used is EL-Nikkor 80mm f/5.6 (old metal version) which has low contrast since it doesn't have coatings as far as I know. I used it for this test since it's commonly used here (for UV photography). 

[photoni]

5 hours ago, LarsHP said:

 

There is absolutely nothing wrong with my monitor. It's an Eizo ColorEdge display, hardware calibrated with X-Rite i1 Display Pro. In other words, it's as close to perfect as it gets. However, the darkest grayscale patch isn't black, because it isn't and shouldn't. 

 

Maybe I could have enhanced contrast further, but the images I have posted have more contrast and saturation than how the ColorChecker looks in real life.

 

The lens used is EL-Nikkor 80mm f/5.6 (old metal version) which has low contrast since it doesn't have coatings as far as I know. I used it for this test since it's commonly used here (for UV photography). 

.

I didn't want to be polemical, I was trying to give you advice on the result you should get.
You have an excellent monitor, an excellent lens, an A7 like mine... the result does not resemble the one proposed on the X-Rite colorchecker site, their reference image is this LINK
I believe that if you underexpose the file and develop the RAW with more contrast you should get closer.
I hope you don't use a LAB or Prophoto color profile, because not even Eizo can reproduce it

.

this is the large version, which is the same as the small one (I have both)

 

[LarsHP]

18 hours ago, photoni said:

.

I didn't want to be polemical, I was trying to give you advice on the result you should get.
You have an excellent monitor, an excellent lens, an A7 like mine... the result does not resemble the one proposed on the X-Rite colorchecker site, their reference image is this LINK
I believe that if you underexpose the file and develop the RAW with more contrast you should get closer.
I hope you don't use a LAB or Prophoto color profile, because not even Eizo can reproduce it

.

this is the large version, which is the same as the small one (I have both)

...

 

It would be interesting if you did a test like I did and see how the colors come out. In sunlight, use the daylight in-camera white balance preset and take a picture with your QB39 1.5mm filter. That filter is not the same as the one I have, which is supposed to be an BG38 equivalent. Actually, Jason says that the QB21 and ICF is the same, even though Tangsinuo has published graphs that show they are different. I guess he means they can be used for the same purpose. As you most likely are aware of, the TSN575 isn't used for normal colors, and is more like a Schott S8612 equivalent, even though it doesn't pass nearly as deep into UV.

 

Regarding the image you posted above, it is way, way too black to my eyes. It doesn't look like reality. (It's also a composite, with the chart photographed separately and added to the image of the woman.) So, whether the image is worthy as a reference, it's a matter of taste in my opinion. That said, even the last series I posted should have lowered the black level, I will give you that. Regarding color balance, it doesn't matter, though.

[LarsHP]

Another strong pointer regarding which of the three thicknesses is closest to the original hot mirror filter is the white balance values when defining neutral in Adobe Camera Raw:

 

2mm ICF:               4400K / -36 tint (green)

3mm QB21:            5500K /  -4 tint (green)

4mm (ICF+QB21): 6800K / +30 tint (magenta)

 

Since we know daylight is around 5400-5600 Kelvin, it is obvious that the two and four millimeter thicknesses are too little and too much respectively, and that 3mm QB21 is quite close. Also, the tint values indicate that the 3mm is close, while the two other thicknesses are off.

[LarsHP]

Here's the 3mm QB21 shot again with white balance set in ACR and black level set correctly, so there is actually zero in at least one of the channels in the blackest spot I could find. While the gray and lighter patches are neutral, the darkest patch has a red tint. I expect this is because there is no dichroic UV and IR cutting filter, which a stock camera has.

[ulf]

2 hours ago, LarsHP said:

(It's also a composite, with the chart photographed separately and added to the image of the woman.) So, whether the image is worthy as a reference, it's a matter of taste in my opinion. That said, even the last series I posted should have lowered the black level, I will give you that. Regarding color balance, it doesn't matter, though.

I do not think the image is a composite. Why do you say that?

If you had followed Tony's link https://calibrite.com/product/colorchecker-classic/ to Calibrite (former x-rite), you would have seen a second photo from what I think was the photo session for the first photo, clearly showing the woman actually holding a calibration target. They come in different sizes.

I agree that Tony's image is too dark and yours are processed a bit too bright and that it does not matter for the colour balance.

There is no material available with zero reflectance and the black patch of an illuminated test target must not be completely black.

My guess is that on these types of targets the reflectance of the black patch should be in the ballpark of 2-4%

[LarsHP]

2 minutes ago, ulf said:

I do not think the image is a composite. Why do you say that?

If you had followed Tony's link https://calibrite.com/product/colorchecker-classic/ to Calibrite (former x-rite), you would have seen a second photo from what I think was the photo session for the first photo, clearly showing the woman actually holding a calibration target. They come in different sizes.

I agree that Tony's image is too dark and yours are processed a bit too bright and that it does not matter for the colour balance.

There is no material available with zero reflectance and the black patch of an illuminated test target must not be completely black.

My guess is that on these types of targets the reflectance of the black patch should be in the ballpark of 2-4%

If you enlarge the model photo and scrutinize how the fingertips look, particularly her left-hand fingertips, I think they look cut to me. Also, the chart itself looks too perfectly lit compared to how the light is in the overview image. There are no signs of shadow and no highlight edges on the chart, which should be present with that lighting. I do think she is holding the chart, but we don't see the chart as she actually held it. The chart is shot separately and added to the scene. That's my best estimation, at least.

[ulf]

I do not agree.

Just for the fun of it I downloaded the 1000x1000 pixel image and manipulated the levels to make the black patch more visible.
There are the highlights and shadows just as I expected at the edges of the patch from the big light-box above and diffuse reflector below infront of the woman.

Maybe we are just seeing what we expect to see and have different expectations. :-)

 

This is not important but interesting to see if there are evidence of a composite or not.

[LarsHP]

14 minutes ago, ulf said:

I do not agree.

Just for the fun of it I downloaded the 1000x1000 pixel image and manipulated the levels to make the black patch more visible.
There are the highlights and shadows just as I expected at the edges of the patch from the big light-box above and diffuse reflector below infront of the woman.

Maybe we are just seeing what we expect to see and have different expectations. :-)

 

This is not important but interesting to see if there are evidence of a composite or not.

 

There are no signs of shadow / highlight except for the black patch, so even based on your image as well, we still disagree.

In addition to my previous remarks, I don't see any structure in the panel or patches, which a real ColorChecker should have.

However, I do recognize that I do not know for sure. I have worked several years in a commercial photo studio, so I may be "colored" by that. 

 

[LarsHP]

23 hours ago, photoni said:

.

I didn't want to be polemical, I was trying to give you advice on the result you should get.
You have an excellent monitor, an excellent lens, an A7 like mine... the result does not resemble the one proposed on the X-Rite colorchecker site, their reference image is this LINK
I believe that if you underexpose the file and develop the RAW with more contrast you should get closer.
I hope you don't use a LAB or Prophoto color profile, because not even Eizo can reproduce it

.

this is the large version, which is the same as the small one (I have both)

 

 

The image you shared (above) has been edited and is not like the original you link to. Dark tones disappear into a black void, contrast is higher and saturation is unrealistically high.

[Andy Perrin]

10 minutes ago, LarsHP said:

I don't see any structure in the panel or patches, which a real ColorChecker should have

Those patches are much larger than a normal colorChecker like yours, so the ratio of the width of each color patch to spacing of the tiny bumps in the patches is much bigger here. At the resolution of the photo, they are probably sub-pixel sized and therefore don’t show up. I agree with those who say it’s not a composite. 
 

Going back to the original discussion, it doesn’t follow that because a certain filter/thickness combination gives you the same colors as the Daylight white balance that those colors are correct. It DOES suggest you have found a decent replacement for the original filter glass.
 

I think to get correct colors you need to do something like make a Custom Light Profile in PhotoNinja (or the equivalent with other software) to map the colors in the RAW to the correct ones. 

[LarsHP]

1 minute ago, Andy Perrin said:

Those patches are much larger than a normal colorChecker like yours, so the ratio of the width of each color patch to spacing of the tiny bumps in the patches is much bigger here. At the resolution of the photo, they are probably sub-pixel sized and therefore don’t show up. I agree with those who say it’s not a composite. 
 

Going back to the original discussion, it doesn’t follow that because a certain filter/thickness combination gives you the same colors as the Daylight white balance that those colors are correct. It DOES suggest you have found a decent replacement for the original filter glass.
 

I think to get correct colors you need to do something like make a Custom Light Profile in PhotoNinja (or the equivalent with other software) to map the colors in the RAW to the correct ones. 

 

As I said from the beginning, the point with this thread is that the 2mm ICF (or QB21) is too thin for at Sony camera. The rest is added as we have gone along.

 

The white balance values in ACR confirm this, as I posted above. I repost them here for convenience:

2mm ICF:               4400K / -36 tint (green)

3mm QB21:            5500K /  -4 tint (green)

4mm (ICF+QB21): 6800K / +30 tint (magenta)

 

I am not sure the IR leak in the 2mm shot would disappear fully, if I created a custom camera profile, because the software only works with the patches, and these don't reflect as much IR as black cloth.

[photoni]

I have two Sony A7s, one Full Spectrum, and one standard (backup if the other one breaks... like the first one I got)
I don't have the QB21, when I want a VIS comparison photo, I settle for the QB39 1.5 mm/t. It has a slight loss in the IR and a more marked loss in the UV but it seems acceptable to me.
- Soligor-KA 35 mm lens @f11

- the white background is 3 mm Teflon, in front there is a plumber's Teflon tape
- late afternoon sun was variable, but exposures are similar.

- Capture One Develop software

- THE ORIGINAL SONY A7 FILTER HAS A THICKNESS OF 1.15 MM
If used with a Hoya R72 it lets a little IR through like the QB39

.

[LarsHP]

Thanks for sharing!

 

It appears your QB39 may be better suited for Sony cameras than the QB21 and ICF filters I have, at least based on this comparison.

As expected, the TSN575 cuts way too much red.

 

What white balance values does Capture One show with your a7FS + QB39 versus the stock camera, when you set neutral in post process?

(It appears that the a7FS+QB39 image has some shadow cast over it / is darker than the standard a7 image, but it is still a good pointer.)