An LR view of the Stouffer wedge

[ErikKaffehr]

Hi,

I have a small write up on observations of a 41 step Stouffer wedge photographed by a Sony Alpha 900. The article is here:

http://echophoto.dnsalias.net/ekr/index.php/photoarticles/56-an-lr-view-of-the-stouffer-wedge

The article has download links for the DNG files, one from the Sony Alpha 900 and the other from the Sony Alpha A55 SLT.

Best regards
Erik

[Guillermo Luijk]

You can obtain accurate RAW histograms in log scale by:

- Developing your DNG files with (this extracts RAW data, and stretches it linearly to the 0..65535 range):
  dcraw -v -d -r 1 1 1 1 -4 -T -t 0 file.dng
- Opening the resulting TIFF in Histogrammar, and then switch to RAW mode.
- Set Gamma=1 (it's 2,2 by default)
- Switch to Log mode
- Perhaps you need to change the Bayer pattern (RGBG by default)

Canon 350D RAW histogram:

[ErikKaffehr]

Thanks!

The major issue right now is that I want a program that identifies each step on the grid. I'm most thankful for your suggestions, and will test "histogrammar" ASAP.

Best regards
Erik

You can obtain accurate RAW histograms in log scale by:

- Developing your DNG files with (this extracts RAW data, and stretches it linearly to the 0..65535 range):
  dcraw -v -d -r 1 1 1 1 -4 -T -t 0 file.dng
- Opening the resulting TIFF in Histogrammar, and then switch to RAW mode.
- Set Gamma=1 (it's 2,2 by default)
- Switch to Log mode
- Perhaps you need to change the Bayer pattern (RGBG by default)

Canon 350D RAW histogram:

[Bart_van_der_Wolf]

Hi,

I have a small write up on observations of a 41 step Stouffer wedge photographed by a Sony Alpha 900. The article is here:

http://echophoto.dnsalias.net/ekr/index.php/photoarticles/56-an-lr-view-of-the-stouffer-wedge

Hi Erik,

Just a few observations.

I noticed (unless my eyes are deceiving me) that you had the tonecurve not set to linear. I'd recommend to set it to linear for such evaluations. One can always add tonecurve adjustments later, i.e. tonemapping of a higher dynamic range to fit the output modality.

I saw that the blackpoint is set to 1, it apparently will influence the colorshifts.

It seems that the Whitebalance was not optimal for the highlights, because the R/G/B peaks in the histogram are not aligned. In my experience, the Stouffer stepwedge itself is pretty neutral in color, but you need to do a WB on the combined lightbox, and stepwedge, and lens, anyway.

The 10 stops DR you found is also influenced by the (inevitable) lens and camera internal veiling glare contribution. It will therefore alway be lower than the sensor DR, but closer to a real life scenario. Different lenses will give different optical dynamic range, and some real life scenes will produce a lower DR because there is more bright stray light bouncing around inside the lens and camera than in the stepwedge shot. You can simulate a worst case scenario by not masking the stepwedge on a lightbox, and having the lightbox extend beyond the FOV for the lens used. Better lenses still produce technically better images. Care must be taken in avoiding ambient light mixing with the backlit setup, so the front of the stepwedge setup should be a shielded dark environment, and the light emitted by the stepwedge should not reflect back onto the target because e.g. the operator is wearing a white T-shirt.

Technically you should verify that the background illumination is sufficiently uniform, or calibrate the results for such an non-uniformity, but maybe you checked and it was all well within the tolerances of the 0.10 OD step size. It is usually sufficient to use a non-calibrated Stouffer stepwedge, because it has quite a good quality, and the calibration only adds a document with density values as measured with a calibated densitometer (of the same physical stepwedge). The actual densities do usually not deviate more than 0.01 or 0.02 OD from the theoretical values, at least on the samples I have used.
 
Cheers,
Bart

[ErikKaffehr]

Hi,

Your eyes are not deceiving you. I simply forgot about setting the curve to linear. Regarding the color balance I have tried different color balances and I think I have seen some shift.

The intention I have with the wedge is more to compare digital with negative film. I'm just about leaving for a two week vacation, but I will update the article as soon I'm back according to your advice.

Always nice to hear from you!

Best regards
Erik

Hi Erik,

Just a few observations.

I noticed (unless my eyes are deceiving me) that you had the tonecurve not set to linear. I'd recommend to set it to linear for such evaluations. One can always add tonecurve adjustments later, i.e. tonemapping of a higher dynamic range to fit the output modality.

I saw that the blackpoint is set to 1, it apparently will influence the colorshifts.

It seems that the Whitebalance was not optimal for the highlights, because the R/G/B peaks in the histogram are not aligned. In my experience, the Stouffer stepwedge itself is pretty neutral in color, but you need to do a WB on the combined lightbox, and stepwedge, and lens, anyway.

The 10 stops DR you found is also influenced by the (inevitable) lens and camera internal veiling glare contribution. It will therefore alway be lower than the sensor DR, but closer to a real life scenario. Different lenses will give different optical dynamic range, and some real life scenes will produce a lower DR because there is more bright stray light bouncing around inside the lens and camera than in the stepwedge shot. You can simulate a worst case scenario by not masking the stepwedge on a lightbox, and having the lightbox extend beyond the FOV for the lens used. Better lenses still produce technically better images. Care must be taken in avoiding ambient light mixing with the backlit setup, so the front of the stepwedge setup should be a shielded dark environment, and the light emitted by the stepwedge should not reflect back onto the target because e.g. the operator is wearing a white T-shirt.

Technically you should verify that the background illumination is sufficiently uniform, or calibrate the results for such an non-uniformity, but maybe you checked and it was all well within the tolerances of the 0.10 OD step size. It is usually sufficient to use a non-calibrated Stouffer stepwedge, because it has quite a good quality, and the calibration only adds a document with density values as measured with a calibated densitometer (of the same physical stepwedge). The actual densities do usually not deviate more than 0.01 or 0.02 OD from the theoretical values, at least on the samples I have used.
 
Cheers,
Bart

[RFPhotography]

Erik, I'm a little confused by something you say in your article.  You state that the grey areas of the histogram represent noise.  The LR user guide states that the grey in the histogram occurs when all 3 RGB channels overlap.  Further, and consistent with this, it states that the cyan, yellow and magenta occur when the blue/green, red/green and red/blue channels respectively overlap.  

This is consistent with the white histogram areas in ACR in PS and the grey overlap in the PS histogram (as well as the CMY colours in both).  Given that, it would make sense that you have more grey in the left side of your histo because there's more alignment of the RGB channels in the darker areas but as the graph shows, the right side is where the channels are more misaligned.

Could you clarify your comments about the grey representing noise?

Thanks.

[Bart_van_der_Wolf]

Could you clarify your comments about the grey representing noise?

Hi Bob,

I'm sure Erik will make adjustments when he finds the time for it. You are correct that it's not so much the grey areas, but more the width of the spikes that represents the amount of noise at the respective brightness levels. Of course, when the width of each R/G/B spike is the same, and they are white balanced (no color shift), then the spikes will look grey (R/G/B aligned and added together).

Cheers,
Bart

[bjanes]

Hi Erik,

Just a few observations.

The 10 stops DR you found is also influenced by the (inevitable) lens and camera internal veiling glare contribution. It will therefore alway be lower than the sensor DR, but closer to a real life scenario. Different lenses will give different optical dynamic range, and some real life scenes will produce a lower DR because there is more bright stray light bouncing around inside the lens and camera than in the stepwedge shot. You can simulate a worst case scenario by not masking the stepwedge on a lightbox, and having the lightbox extend beyond the FOV for the lens used. Better lenses still produce technically better images. Care must be taken in avoiding ambient light mixing with the backlit setup, so the front of the stepwedge setup should be a shielded dark environment, and the light emitted by the stepwedge should not reflect back onto the target because e.g. the operator is wearing a white T-shirt.

Excellent points, Bart. Here is the worst case scenario that you mentioned using the Nikon D3 with the 60 mm Micro Nikkor  with the nano lens coating. I don't have test data for flare with this lens, but it should have less than average flare, certainly less than with a zoom.

It is useful to plot the curve rather than merely looking at the histograms. One can see the departure from linearity in the shadows that I presume is mainly due to veiling glare.

Regards,

Bill

[RFPhotography]

Hi Bart.  Yes, that's what I understand as well.  In my crude understanding, the width represents the std devs away from the mean and if all pixels were the same (i.e., no noise) there'd be a single spike at the mean. 

[Guillermo Luijk]

I plotted the RAW histograms in EV divisions. For being a 12-bit camera they are difficult to follow in the deep shadows (just 1 level in -11EV, 2 levels in -10EV,...).

Sony Alpha 900

[ejmartin]

I plotted the RAW histograms in EV divisions. For being a 12-bit camera they are difficult to follow in the deep shadows (just 1 level in -11EV, 2 levels in -10EV,...).

Perhaps of some interest would be the raw histograms after performing a gaussian blur on each channel (I believe one can separate channels using eg libraw).  The eye perceives a coarse-grained image that is the average over the circle of confusion.  A blur at the CoC will not have any visible effect, but will smooth the quantization jumps and probably recover smoother Gaussians in the lower EV regime.  Where the Gaussians merge will indicate the DR for that spatial resolution; doing this for several blur ranges, one can then extrapolate back to get the pixel level DR perhaps a bit more cleanly.

[Guillermo Luijk]

Perhaps of some interest would be the raw histograms after performing a gaussian blur on each channel (I believe one can separate channels using eg libraw).  The eye perceives a coarse-grained image that is the average over the circle of confusion.  A blur at the CoC will not have any visible effect, but will smooth the quantization jumps and probably recover smoother Gaussians in the lower EV regime.  Where the Gaussians merge will indicate the DR for that spatial resolution; doing this for several blur ranges, one can then extrapolate back to get the pixel level DR perhaps a bit more cleanly.

Good idea. I extracted the R channel with PS's 50% nearest neighbour resizing, blurred with 2px radius, and some new gaussians became visible. The already visible got thinner thanks to noise reduction:



I tried other blurring radius obtaining similar results.

[ErikKaffehr]

Hi!

I made some updates to the my web page on the Stouffer edge. The main addition is that I added som analysis made with Imatest. A posting by Bill (bjanes) reminded me of Imatest having support for multistep wedges.

I used Imatest with both dcraw and LR. To my surprise it seems that LR can yield a significantly wider DR than dcraw. The image from LR was in prophoto RGB baut I think the Imatest takes the gamma into account.

Using luminance noise reduction in LR did not affect total DR according to Imatest, but did improve the usable DR.

The images I published do not pretend to be scientific, my setup is not good enough.

Best regards
Erik