HDR 64 Bit, STF Posterization and Linear State Noise Reduction

[dgbarar]

I am performing my first HDR images to reduce blooming.  As everyone is aware, the 64 bit HDR images maybe somewhat to significantly posterized when applying the screen transfer function.  I like to apply some noise reduction (e.g. TGVDenosie) while the image is linear using previews to optimize the denoise parameters.  With the image posterized, it can be difficult to optimize the denoise parameters.

The only method that I have to come up with is to iteratively apply denoiseing, perform a non-linear stretch, and evaluate the image.  However, this is time consuming.  What other techniques should I consider if my 64 bit image shows significant posterization making it difficult optimize denoise parameters.

Thanks in advance for your help.

Don

[dgbarar]

Hi,

Does anyone have thoughts how to best approach noise reduction on 64 bit HDR linear images when they may be moderately to severely posterized when viewed with STF applied.  I recognize that posterization is not present after a non-linear operation. This would be a big help to me.

Thanks again in advance.

Don

[naavis]

I've been wondering the same thing. Also, if I convert a 64-bit HDR image into 32-bit with SampleFormatConversion, the image _still_ posterizes. If I do a HistogramTransformation with the same settings as STF, the data is not posterized, so the fault must lie with STF. I'm wondering if this is a bug.

[Juan Conejero]

There's no bug; it is a limitation. The problem is that screen transfer functions work on 16-bit histograms, and 65536 discrete values are clearly insufficient for HDR images, especially for very deep HDR compositions. So the problem is just the screen representation; of course when you apply the HistogramTransformation tool there is no posterization because all processing tools work with the actual image data.

During the 1.8 cycle (although not in the initial 1.8.0 release) we'll have optional 20 and 24-bit histograms and much better and efficient image statistics in the PixInsight Core application. This will fix all of these screen rendering problems. In the meantime, I suggest making temporary stretched images with HT. I'm sorry, I have no better solution to this problem right now.