HDRI capture of LED (histograms and range)

Hello!

I've been experimenting with the number of photos to include in final HDRI.

I took a sequence of photos of a single LED with reflectance standards
included in the scene.

EOS7D 28-105mm lens at 28mm

F16 1/8000-1/15’’

F4 1/30-5 mins

With the ND filter t=0.0094

Images are fused with raw2hdr.

They were calibrated at white reflectance standard 215 cd/m2.

I’ve noticed the following tendency:

If I fuse different number of photos (cut the number of photos on the
shortest end), after calibrating at white reflectance standard, I get
different luminance values for the LED.

*Shortest exposure to fuse L,cd/m2*

1/125’’ f16 4.5*106

1/250’’ f16 9.06*106

1/500’’ f16 18*106

I've seen interesting discussion between Axel and Greg on photos to
include. But it seems like there are many uncertainties.

In HDRI second edition book it says "The darkest exposure should have no
RGB values greater than 200 or so, and the lightest exposure should have no
RGB values less than 20 or so. Do NOT include an excess of exposures beyond
this range, as it will do nothing to help with the response recovery and
may hurt."
I assume it is the same for any HDRI sequence, not only for response curve.

I have plenty of photos of dark exposures that have no values greater than
200, same with the light exposures and 20.If somebody can clarify what
photos should be included or have any other suggestions that would be great

Thank you,
*Yulia *

Hi Yulia,

Since your maximum values are tracking the differences in your shorter exposures, it seems to indicate that you still haven't captured the brightest point on the LED. What are you looking at that makes you think the exposures have nothing over 200?

There is no harm in the newer version of hdrgen in including exposures that are too dark or too light. More naive methods might have trouble with this, but not hdrgen or raw2hdr.

Best,
-Greg

···

From: "Tyukhova, Yulia" <[email protected]>
Date: March 1, 2012 12:46:29 PM PST

Hello!

I've been experimenting with the number of photos to include in final HDRI.

I took a sequence of photos of a single LED with reflectance standards included in the scene.
EOS7D 28-105mm lens at 28mm
F16 1/8000-1/15’’
F4 1/30-5 mins
With the ND filter t=0.0094
Images are fused with raw2hdr.
They were calibrated at white reflectance standard 215 cd/m2.
I’ve noticed the following tendency:
If I fuse different number of photos (cut the number of photos on the shortest end), after calibrating at white reflectance standard, I get different luminance values for the LED.
Shortest exposure to fuse L,cd/m2
1/125’’ f16 4.5*106
1/250’’ f16 9.06*106
1/500’’ f16 18*106

I've seen interesting discussion between Axel and Greg on photos to include. But it seems like there are many uncertainties.

In HDRI second edition book it says "The darkest exposure should have no RGB values greater than 200 or so, and the lightest exposure should have no RGB values less than 20 or so. Do NOT include an excess of exposures beyond this range, as it will do nothing to help with the response recovery and may hurt."
I assume it is the same for any HDRI sequence, not only for response curve.

I have plenty of photos of dark exposures that have no values greater than 200, same with the light exposures and 20.If somebody can clarify what photos should be included or have any other suggestions that would be great

Thank you,
Yulia

I suggest to use two parallel approaches, one with your LED, one with an incandescent low voltage. The problem with LED luminance measurements from HDRI is that a) the color gamut is small and b) the photopic sensitivity curve V(lambda) currently used is incorrect. It underestimates blue wavelengths quite a bit. Therefore, start with an incandescent first, because LEDs have a strong blue peak around 440 or 460 nm, depending on the manufacturer. You are dealing with the liumitations of V(lambda) and then with tristimulus values which are just not good enough for narrowband LEDs, including white ones with a blue peak.

When you use the incandescent instead use it in a black lab. Place a few white paper samples on the walls at different known locations. Place your incandescent at a defined location. Derive the geometry, all distances and all angles between those white papers and the lamp. Turn the incandescent on and the room lights off. Measure the luminance of the white papers at a location that you marked with a soft pencil. Make HDR images and determine the luminance and scale.

Based on the luminance values of the papers, you can now derive the luminance of the incandescent filament. Make HDR images of the filament as seen from those papers. Check if the corresponding illuminance values of the white paper at reflectance around 85% corresponds to the luminance values of the filament in that direction.

Repeat several times to give you confidence.

Once you are heading in the right direction, you can repeat the whole procedure with an LED. But this time your values will be around 20-30% off. You should really use a CCD camera with a photopic filter to get better values, and then try to come up with a procedure to minimize errors for the HDR approach.

Regards

Martin Moeck
Osram

···

________________________________
From: Tyukhova, Yulia [mailto:[email protected]]
Sent: Thursday, March 01, 2012 9:46 PM
To: [email protected]
Subject: [HDRI] HDRI capture of LED (histograms and range)

Hello!

I've been experimenting with the number of photos to include in final HDRI.

I took a sequence of photos of a single LED with reflectance standards included in the scene.
EOS7D 28-105mm lens at 28mm
F16 1/8000-1/15''
F4 1/30-5 mins
With the ND filter t=0.0094
Images are fused with raw2hdr.
They were calibrated at white reflectance standard 215 cd/m2.
I've noticed the following tendency:
If I fuse different number of photos (cut the number of photos on the shortest end), after calibrating at white reflectance standard, I get different luminance values for the LED.
Shortest exposure to fuse L,cd/m2
1/125'' f16 4.5*106
1/250'' f16 9.06*106
1/500'' f16 18*106

I've seen interesting discussion between Axel and Greg on photos to include. But it seems like there are many uncertainties.

In HDRI second edition book it says "The darkest exposure should have no RGB values greater than 200 or so, and the lightest exposure should have no RGB values less than 20 or so. Do NOT include an excess of exposures beyond this range, as it will do nothing to help with the response recovery and may hurt."
I assume it is the same for any HDRI sequence, not only for response curve.

I have plenty of photos of dark exposures that have no values greater than 200, same with the light exposures and 20.If somebody can clarify what photos should be included or have any other suggestions that would be great

Thank you,
Yulia

Hi Yulia and all,

I just have one small question for understanding what you are doing. Did you derive the response curve of your camera using another setup (not the directly visible LED) and reuse that camera response later on your LED captures? Or do you take the camera response from the same set of images (showing the LED in an otherwise dark room) that you are going to assemble into a HDR?

Cheers, Lars.

Hello everybody!

I appreciate your help and responses.

Greg,

I've been checking histograms with CANON Digital Photo Professional. I was
checking L channel, I guess I need to make sure RGB values are below 200 as
well. Are there any suggestions on this issue?

Martin,

That is a very interesting suggestion! I guess I can see the difference
between the measurements of two light sources.

..the photopic sensitivity curve V(lambda) currently used is incorrect.

It underestimates blue wavelengths quite a bit.

You should really use a CCD camera with a photopic filter to get better

values..

You say that V(lambda) underestimates the values. But doesn't CCD camera's
photopic filter have the same response curve? So, it would have the same
mistake.

Lars,

Did you derive the response curve of your camera using another setup

(not the directly visible LED) and reuse that camera response later on your LED
captures? Or do you take the camera response from the same set of images
(showing the LED in an otherwise dark room) that you are going to assemble
into a HDR?

I've tried two approaches. The first one is to derive the response curve
from the scene with very dark and bright areas with smooth gradients
and neutral colors. Then fuse them to obtain RC. And then I've used it in
photosphere for subsequent HDRIs. And the second one is to fuse raw images
with Greg's script raw2hdr, where you don't have to have response curve.
Greg: "The raw2hdr script doesn't need to derive a response curve, since
the sensor data is linear. Instead, it creates an output from dcraw that
follows a 2.0 gamma and creates an artificial response curve of x^2 to
decode it. This reduces quantization errors from the 8-bit intermediate
images". And then I can analyse my hdr image in Photosphere.

···

--
Thank you,
*Yulia Tyukhova*
*
*
Fulbright Scholar, "Intern LC"
Architectural Engineering Graduate Student, UNL-Omaha, NE, USA
B.E. and M.E. in Lighting Engineering (MPEI), Moscow, Russia
[email protected]
[email protected]
+1 (402) 996 0910
PKI 247

Hi Yulia,

From: "Tyukhova, Yulia" <[email protected]>
Date: March 2, 2012 9:14:15 AM PST

Greg,

I've been checking histograms with CANON Digital Photo Professional. I was checking L channel, I guess I need to make sure RGB values are below 200 as well. Are there any suggestions on this issue?

Photosphere or Photoshop can give you an RGB histogram (along with max. values) from an 8-bit image.

Martin,

That is a very interesting suggestion! I guess I can see the difference between the measurements of two light sources.

>> ..the photopic sensitivity curve V(lambda) currently used is incorrect. It underestimates blue wavelengths quite a bit.
>> You should really use a CCD camera with a photopic filter to get better values..

You say that V(lambda) underestimates the values. But doesn't CCD camera's photopic filter have the same response curve? So, it would have the same mistake.

Typical cameras do not follow the CIE standard observer curves for a number of reasons. Instead, they use red, green and blue bandpass filters and a color matrix optimized for color reproduction using some set of patches. It's an under-constrained problem, and different makers will optimize their color transform matrix differently. This is why dcraw tends to be more reliable -- Dave Coffin always derives his CTM the same way and doesn't bias it towards one set of colors or another.

Even so, Martin is correct that highly saturated colors, such as those produced by typical LEDs, will often cause problems for reproduction and luminance estimation. There is no easy fix for this, unfortunately.

Best,
-Greg

Another thing to note on incandescent lamps. Probably applies to none but me….
But one thing to do is to check what is the sensitivity of your camera to the infrared.
For example the Leica M8 is well known to be extremely sensitive, to the point of requiring filters to cut UV/IR.
And this is usually visible in the black fabrics which become purple.
If so use filters or expect to measure higher luminance values when IR is present.

This also applies to the old Nikon D100 and probably other 2005ish cameras
G

···

On 2 Mar 2012, at 12:04, Moeck, Dr. Martin wrote:

I suggest to use two parallel approaches, one with your LED, one with an incandescent low voltage. The problem with LED luminance measurements from HDRI is that a) the color gamut is small and b) the photopic sensitivity curve V(lambda) currently used is incorrect. It underestimates blue wavelengths quite a bit. Therefore, start with an incandescent first, because LEDs have a strong blue peak around 440 or 460 nm, depending on the manufacturer. You are dealing with the liumitations of V(lambda) and then with tristimulus values which are just not good enough for narrowband LEDs, including white ones with a blue peak.

When you use the incandescent instead use it in a black lab. Place a few white paper samples on the walls at different known locations. Place your incandescent at a defined location. Derive the geometry, all distances and all angles between those white papers and the lamp. Turn the incandescent on and the room lights off. Measure the luminance of the white papers at a location that you marked with a soft pencil. Make HDR images and determine the luminance and scale.

Based on the luminance values of the papers, you can now derive the luminance of the incandescent filament. Make HDR images of the filament as seen from those papers. Check if the corresponding illuminance values of the white paper at reflectance around 85% corresponds to the luminance values of the filament in that direction.

Repeat several times to give you confidence.

Once you are heading in the right direction, you can repeat the whole procedure with an LED. But this time your values will be around 20-30% off. You should really use a CCD camera with a photopic filter to get better values, and then try to come up with a procedure to minimize errors for the HDR approach.

Regards

Martin Moeck
Osram
From: Tyukhova, Yulia [mailto:[email protected]]
Sent: Thursday, March 01, 2012 9:46 PM
To: [email protected]
Subject: [HDRI] HDRI capture of LED (histograms and range)

Hello!

I've been experimenting with the number of photos to include in final HDRI.

I took a sequence of photos of a single LED with reflectance standards included in the scene.
EOS7D 28-105mm lens at 28mm
F16 1/8000-1/15’’
F4 1/30-5 mins
With the ND filter t=0.0094
Images are fused with raw2hdr.
They were calibrated at white reflectance standard 215 cd/m2.
I’ve noticed the following tendency:
If I fuse different number of photos (cut the number of photos on the shortest end), after calibrating at white reflectance standard, I get different luminance values for the LED.
Shortest exposure to fuse L,cd/m2
1/125’’ f16 4.5*106
1/250’’ f16 9.06*106
1/500’’ f16 18*106

I've seen interesting discussion between Axel and Greg on photos to include. But it seems like there are many uncertainties.

In HDRI second edition book it says "The darkest exposure should have no RGB values greater than 200 or so, and the lightest exposure should have no RGB values less than 20 or so. Do NOT include an excess of exposures beyond this range, as it will do nothing to help with the response recovery and may hurt."
I assume it is the same for any HDRI sequence, not only for response curve.

I have plenty of photos of dark exposures that have no values greater than 200, same with the light exposures and 20.If somebody can clarify what photos should be included or have any other suggestions that would be great

Thank you,
Yulia
_______________________________________________
HDRI mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/hdri

I suggest to use two parallel approaches, one with your LED, one with
an incandescent low voltage. The problem with LED luminance
measurements from HDRI is that a) the color gamut is small and b)
that the photopic sensitivity curve V(lambda) currently used is
incorrect. It underestimates blue wavelengths quite a bit. Therefore,
start with an incandescent first, because LEDs have a strong blue peak
around 440 or 460 nm, depending on the manufacturer. Therefore, you
have to mess with V(lambda) and then with tristimulus values which are
just not good enough for LEDs, including white ones.

When you use the incandescent instead use it in a black lab. Place a
few white paper samples on the walls at different known locations.
Place your incandescent at a defined location. Derive the geometry,
all distances and all angles between those white papers and the lamp.
Turn the incandescent on and the room lights off. Measure the
luminance of the white papers at a location that you marked with a
soft pencil. Make HDR images and determine the luminance and scale.

Based on the luminance values of the papers, you can now derive the
luminance of the incandescent filament. Make HDR images of the
filament as seen from those papers. Check if the corresponding
illuminance values of the white paper at reflectance around 85%
corresponds to the luminance values of the filament in that direction.

Repeat several times to give you confidence.

Once you are heading in the right direction, you can repeat the whole
procedure with an LED. But this time your values will be around 20-30%
off. You should really use a CCD camera with a photopic filter to get
better values, and then try tom come up with a procedure to minimize
errors for the HDR approach.

Regards

Martin Moeck
Osram

···

On Thu, Mar 1, 2012 at 9:46 PM, Tyukhova, Yulia <[email protected]> wrote:

Hello!

I've been experimenting with the number of photos to include in final HDRI.

I took a sequence of photos of a single LED with reflectance standards
included in the scene.

EOS7D 28-105mm lens at 28mm

F16 1/8000-1/15’’

F4 1/30-5 mins

With the ND filter t=0.0094

Images are fused with raw2hdr.

They were calibrated at white reflectance standard 215 cd/m2.

I’ve noticed the following tendency:

If I fuse different number of photos (cut the number of photos on the
shortest end), after calibrating at white reflectance standard, I get
different luminance values for the LED.

Shortest exposure to fuse L,cd/m2

1/125’’ f16 4.5*106

1/250’’ f16 9.06*106

1/500’’ f16 18*106

I've seen interesting discussion between Axel and Greg on photos to include.
But it seems like there are many uncertainties.

In HDRI second edition book it says "The darkest exposure should have no RGB
values greater than 200 or so, and the lightest exposure should have no RGB
values less than 20 or so. Do NOT include an excess of exposures beyond this
range, as it will do nothing to help with the response recovery and may
hurt."
I assume it is the same for any HDRI sequence, not only for response curve.

I have plenty of photos of dark exposures that have no values greater than
200, same with the light exposures and 20.If somebody can clarify what
photos should be included or have any other suggestions that would be great

Thank you,
Yulia

_______________________________________________
HDRI mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/hdri