Deprecated: preg_replace(): The /e modifier is deprecated, use preg_replace_callback instead in /www/htdocs/w0068b6d/smf/Sources/Load.php(225) : runtime-created function on line 3
How to create a TRC (Tone Reproduction Curve) using version 2 of FDRTools?
forum@fdrtools.com
June 20, 2019, 09:04:42 AM *
Welcome, Guest. Please login or register.

Login with username, password and session length
News: SMF - Just Installed!
 
   Home   Help Search Login Register  
Pages: [1]
  Print  
Author Topic: How to create a TRC (Tone Reproduction Curve) using version 2 of FDRTools?  (Read 16805 times)
ericob
New
*
Posts: 11


View Profile
« on: November 14, 2007, 09:27:53 PM »

My input files are 16 bit TIFFs (stitched panoramas).  It seems I am getting fine results when using a TRC setting of "none."  The tutorial on how to create a TRC refers to a feature that does not seem to be in version 2 of FDRTools.

How important is it to use an accurately-created TRC?  And is this a "camera" reproduction curve, or more like a "system" reproduction curve? 

For example, when I convert my original raw files to TIFFs (in preparation for stitching) I have, as much as possible "zeroed" or "neutralized" nearly all of the adjustments available (I use Adobe Lightroom).  If I change the color temperature, I make the same adjustment for all the images in all three bracketed sets. 

However, if I decided to switch to always setting "vividness" to 30%, I'm guessing creating a TRC from images like that would produce a different result.

Thanks!
Logged
AndreasSchoemann
Administrator
Resident
*****
Posts: 551


View Profile Email
« Reply #1 on: November 15, 2007, 09:44:03 AM »

My input files are 16 bit TIFFs (stitched panoramas).  It seems I am getting fine results when using a TRC setting of "none."  The tutorial on how to create a TRC refers to a feature that does not seem to be in version 2 of FDRTools.

Unfortunately not. In the course of restructuring the Gui I switched this functionality off. Wanted to reintegrate it with 2.0 in a modified, more intuitively usable form but time did not allow to do it. So you need to produce TRCs with 1.8.x (different versions of FDRTools can be installed and used in parallel).

Quote
How important is it to use an accurately-created TRC?  And is this a "camera" reproduction curve, or more like a "system" reproduction curve?

It is not mandatory to get fine results. It is not necessary at all if you read RAWs directly. It is useful if you read camera JPEGs.

The motivation behind this kind of calibration is (as always when you do calibration stuff) to standardize measurement results for further processing. In this particular case the reasoning behind creating a TRC is to linearise the input images as good as possible. Nonlinear input is not good for HDR merging. It forces you to use tricks to blend the images seamlessly. Look at the following images to see what I mean:



  • To the left you can see the result when reading RAWs. See how the histograms of the exposures line up nicely. This is because the RAW data is linear by nature, it is not "manipulated". The resulting HDR image is simply gathered by adding up the contributions of each of the layers (dark toned areas), see the HDR image histogram above. This is the "perfect" HDR image.
  • The middle image shows the camera JPEGs. See the histograms. They have different width and don't line up well. In fact it looks like chaos, compared to the RAWs. And that is despite the fact that they are already linearised using the standard TRC included in the sRGB color profile! But as you can see the linearisation is poor.
  • The right one shows JPEGs, linearised with a custom TRC (it is the "Canon D60 JPEG" TRC I've created for my camera). Compare the histograms. The histograms line up much better compared to the sRGB linearised JPEGs. The resulting HDR image is much closer to the RAW version, just compare the HDR image histograms.

FDRTools yields fine results with camera JPEGs too. The necessary trickery to blend JPEGs seamlessly is done behind the scenes. No need to worry about that. However, merging RAWs to HDR results in higher quality images. You can see that in the tone mapped image.

Quote
For example, when I convert my original raw files to TIFFs (in preparation for stitching) I have, as much as possible "zeroed" or "neutralized" nearly all of the adjustments available (I use Adobe Lightroom).  If I change the color temperature, I make the same adjustment for all the images in all three bracketed sets. 

However, if I decided to switch to always setting "vividness" to 30%, I'm guessing creating a TRC from images like that would produce a different result.

If I knew exactly what "vivdness" does I could answer this. Probably yes. I think Erik Krause has done an investigation on how the ACR parameters influence the linearity of the resulting developped images. Search the http://wiki.panotools.org/ for details.

Your best bet is probably to apply a linear color profile to the developed images in ACR, though I'm not sure if this this meanwhile possible. It is not in CS2.

In general you create a TRC under certain assumptions. You will only get good results if your images hold these assumptions. E.g. if "vividness" changes the curves of your images (which is probably the case) then applying TRC done with different "vividness" gives different results. These may still be far better than not applying a custom TRC. You have to test it. Another important assumption is to be sure that the camera does not apply "magic" in-camera adjustments to your images, depending on camera settings like White Balance etc. If that's the case then your TRC will probably fail most often. The D60 behaves well in this respect. Again you have to test that. The image histograms in FDRTools help you in judging the quality of linearisation.

Hmmmh, I hope I haven't forgotten anything. If you have further questions...

[Have just seen that the images are wrapped if your browser window is not wide enough. Will see if I can configure the forum not to do this.][No, does not seem to be possible.]
« Last Edit: November 15, 2007, 10:41:53 AM by AndreasSchoemann » Logged
ericob
New
*
Posts: 11


View Profile
« Reply #2 on: November 19, 2007, 10:40:24 PM »

(I see you must have a wide screen Smiley !  Conveniently, I recently bought one also and with the browser window expanded to the full width of the screen I can indeed view the screen shots side by side.)

Thanks for the tip about the previous version still being available... I hadn't noticed.

Special thanks for the detailed information about Tone Reproduction Curves!

Since I am working with images several steps away from the raw file, it was helpful that you confirmed my suspicion that any step in that process that affects the color curves will change the TRC that results from following your tutorial. 

The histograms for my images appear relatively the same.  I hope this is because the images have something like linear data in them.  It could be that there are discontinuities in the histograms that are masked by the fact that the source images are 6000 x 3000 16-bit TIFFs.  I'll have to go the the effort, some time, of creating a TRC and then see if my output images look much different.

eo
Logged
AndreasSchoemann
Administrator
Resident
*****
Posts: 551


View Profile Email
« Reply #3 on: November 20, 2007, 12:42:21 PM »

(I see you must have a wide screen Smiley !  Conveniently, I recently bought one also and with the browser window expanded to the full width of the screen I can indeed view the screen shots side by side.)

Yes, that's true. Wide screens are convenient indeed :-)

Quote
The histograms for my images appear relatively the same.  I hope this is because the images have something like linear data in them.  It could be that there are discontinuities in the histograms that are masked by the fact that the source images are 6000 x 3000 16-bit TIFFs.

This is important: if you feed your source images to a stitching program then the resulting panorama is not usable for proper linearisation any more. Reason is: the blending process done via enblend, smartblend or other algorithms modifies pixel intensities locally. Local modifications of any kind (another example is local tone mapping) make it virtually impossible to recover the true (unmodified) pixel intensity values, i.e. the sensor data.

Hence: TRC should only be applied to JPEGs that come straight from the camera or to TIFFs that have undergone repeatable modifications. Things that come to my mind: vignetting correction and TCA correction (which are calibrations also) and maybe applying identical "vividness" to all images.

Logged
Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.11 | SMF © 2006-2009, Simple Machines LLC Valid XHTML 1.0! Valid CSS!