Inconsistent HDR composition detected

[EelkoGielis]

Hi all,

When I use the HDRComposition tool I get the next message:

* HDR image component 1 of 3
Reading TIFF: 32-bit floating point, 3 channel(s), 6080x4044 pixels, chunky: 100%

* Initial quantization:
q0 = 2.59236e+006 (21.31 bits)
q1 = 1.08073e+006 (20.04 bits)
q2 = 1.85786e+006 (20.83 bits)

* Integrating HDR image component 2 of 3
Reading TIFF: 32-bit floating point, 3 channel(s), 6080x4044 pixels, chunky: 100%
Building HDR composition mask: 100%
Fitting linear HDR image: |
*** Error: Inconsistent HDR composition detected (bad linear fit, channel 0)

I loaded 32/bit color TIF images, just made in PI 1.7 to do a ´fake´ HDR.

Is there something I can do without degrading my input images?

[georg.viehoever]

How did you process the images before this? Are they still linear? Are they calibrated?
Georg

[EelkoGielis]

I post-processed Nikon NEF files in DeepSkyStacker with calibration of darks, flats and bias frames.

The output is a 32-bit floating point TIF image.

I then used a HistogramTransformation in the mid tones of 0.4 for 12 times and saved each step as a 32-bit floating point TIF.

I did that before in 1.6 and that worked.

[georg.viehoever]

You need to do HDRComposition before the HistogramTransform. In particular the midtones change is non-linear. Most likely this is the reason why you have the problems.
Georg

[EelkoGielis]

This might sound silly, but how do I stretch an image in a linear way?

[georg.viehoever]

1. Best is not to stretch it at all... Use STF to make something visible on the screen, and delay any actual stretching to afte HDRComposition.
2. You can do it in PixelMath, using a formula such as 20*$T
3. slightly unsave: Move only the highlights slider in histogram transform
Georg

[EelkoGielis]

Hi Georg,

I got it working now, see this image:

http://www.flickr.com/photos/14721988@N02/8395284913/

The workflow is under the image.

-Eelko.

[Andres.Pozo]

Hi Georg,

I got it working now, see this image:

http://www.flickr.com/photos/14721988@N02/8395284913/

The workflow is under the image.

-Eelko.

You keep stretching the images before HDRComposition. You can not do that if you want a good result from HDRComposition. It is designed to work with lineal images and applying a HT with a midtones value other that 0.5 destroys the linearity.

The workflow must be:

• Calibration of the subs with at least darks and flats. This is MANDATORY.
• Alignment of all the subs against the same reference.
• Integrate all the images for each exposure.
• Apply HDRComposition.
• Now you can process the image as you like.

Before HDRComposition you can apply transformations that don't affect the linearity, but I don't recommend that since they can produce artifacts after HDRComposition.

Also in HDR images it is usually necessary to apply a HDRMultiscaleTransform step to compress the dynamic range.

[EelkoGielis]

Hi Andres,

The workflow must be:

• Calibration of the subs with at least darks and flats. This is MANDATORY.
• Alignment of all the subs against the same reference.
• Integrate all the images for each exposure.
• Apply HDRComposition.

Now you can process the image as you like.

As you can see, I did calibrate the images I used, I just had the one image. Alignment is done in DeepSkyStacker and I did not have time to get more exposures.

All in all, this is the best I can do with this image.

-Eelko.

[Andres.Pozo]

As you can see, I did calibrate the images I used, I just had the one image. Alignment is done in DeepSkyStacker and I did not have time to get more exposures.

I don't understand. What does it mean that you have only one image? For HDRcomposition you must have at least two images of different exposition time.

All in all, this is the best I can do with this image.

You stretched 12 times the image before HDRComposition. You shouldn't do that.

[EelkoGielis]

Hi Andres,

I stated in the first post that I wanted to do a 'fake' HDR.
The only image I have is a stacked image from 18 x 4m41s long.

If I had had more scope time, I would have done it differently.

-Eelko.

[pfile]

since PI's HDR tool requires linear images, i think you'd be better off using enfuse. enfuse is actually for terrestrial images so it can deal with nonlinear data. i have used that in the past on M42 in exactly the way you are describing. worked pretty well.

[EelkoGielis]

Thanks Rob,

I'll give that a try somewhere this week.