Error in HDR Composition?

[sdh]

Hi, I'm an inexperienced user of Pixinsight, and I'm getting an error when I try HDR Composition. It says "Inconsistent HDR composition detected (bad linear fit, channel 0)". I'm using PixInsight Core 01.07.00.0702 Starbuck (x86_64) on Windows 7.

This error only occurs on certain combinations of data files. All the data files are from MaxIM DL, and they are in what MaxIM calls "32bit IEEE floating point format FITS". I am fairly competent with MaxIM, Images Plus, and CCD Inspector, and this data has not shown any problems outside of HDR Composition.

Essentially, I have 3 luminance data sets. For each set, in MaxIM, I produced a sigma-clip median of the set, and a sum of the set. Then I summed various combinations among the 3 sets. When the problem with HDR Combination occured, I tried various 2-file combinations. Some combinations fail, some succeed.

In looking at my notes, it is difficult to hypothesize what is causing the failure. If I had to guess, I'd say its related to the fact that a sigma-clip median file has relatively low data values, while a summed file has relatively high values. The data does not appear to be clipped by 32bit representation.

I am not sure what my next step should be. If anyone has any information or advise, I would be grateful.

Steve

[Juan Conejero]

Hi Steve,

Most likely the problem is that the 32-bit floating point images are being incorrectly rescaled when you open them with PixInsight. This happens because the FITS standard does not provide any means to specify the numeric range to which floating point pixel values are referred (the black and white points). Since PixInsight and the other application use different default ranges for floating point data, PI decides to rescale what it interprets as out-of-range values.

I see two solutions:

1. Use format hints to force HDRComposition to use the appropriate range of floating point pixel values. In the case of MaxIM:

lower-range 0 upper-range 65535

will probably load the images correctly, since this application seems to store floating point pixel data in the 16-bit integer range [0.0,65535.0]. The above line should be used in the input hints field of HDRComposition.

2. Use PixInsight to integrate your raw frames. Our ImageIntegration tool provides superior performance and much more advanced features. It will generate integrated images in PixInsight's [0,1] native floating point range, so you won't need format hints.

Let me know if this helps.

[sdh]

Hi Juan, thanks for the reply.

After thinking some more about my data set, and about Rogelio's tutorial on HDRComposition with PixInsight, I realized that the full sum of my luminance data could proceed directly to HDRWavelet processing. This is because none of my 60 subexposures are over-exposed (>16bits), and because the sum of all 60 does not exceed the 32bit floating-point capability of MaxIM.

I would like to try your suggestions, but I may not be able to until the weekend. I shall post the results.

Regarding my results with HDRWavelets, I am unclear whether they are are limited by my small skill with PixInsight, or limited by the quality of my data. I would be happy to upload it for someone else to attempt processing, but my guess is no one has the time to do so  Perhaps some day you could hire interns and provide this on a per-fee basis ... but I am no expert on business modelling either, hahahahah ...

Btw, as a software engineer myself, I have wondered if some kind of false-color strategy could be applied to allow a user to gauge the quality of their raw data. However the idea has never gone beyond that, again, due to time limitations.

Cheers,
Steve

[sdh]

Hi Juan, just so you know, I tried the format hints you suggested, but I still got the message about
"bad linear fit." However this is not an issue for me because none of my subexposures are saturated.

I've moved on to using HDRWavelets, and my results are improving as I become more familiar with it.
I have found it helpful to systematically vary 1 parameter at a time, save each result as a descriptively
named jpg, then page thru the jpgs and try to characterize how the parameter changes the image.
Aligning and integrating subexposures using PixInsight works fine, and I am migrating towards that.

As always, taking notes helps :-)

Steve

[pfile]

i was just messing with this myself and i think that the bad linear fit error happens if the default binarizing threshold happens to be too high for the data you've got. i think this message sort of means "there was no data from this particular image, so when i tried to do a linear fit, it failed"