Another update:
This is a scene where the HRSC nadir channel is used to pan-sharpen the colourised MOC data (corresponding to the MOC+HRSC images in my previous post):
This is the same scene, now using the nadir channel to pan-sharpen the actual HRSC colour channels, before low frequency-matching them to the MOC colour reference:
The HRSC colour data definitely make the scene more dramatic (and hopefully provide higher frequency colour information), not to mention replacing fake colour data with real ones (at least at high spatial freqencies).
I noticed some slight striping artefacts in the HRSC colour channels, but hopefully they won't be apparent close up in Orbiter. At the lower resolution levels I might gradually fade out the HRSC colours.
I've got one more idea up my sleeve I want to try before starting on the global mosaic in earnest: Currently I am matching the low resolution colour reference by rescaling the low freqency band of the RGB channels to match the means of the reference. While this matches the mean intensity, it doesn't match the contrast. This can be a problem if the brightness difference between HRSC fragment and reference data is high. For example, if the HRSC sample is much darker than the reference, rescaling the means might lead to a much higher contrast than the reference.
To address this, I am going to try to match the low-frequency variance as well as the mean. This would match variance multiplicatively, and mean additively. Of course, this operation might subdue the drama of the above scene, so I'll have to see how it looks. Anyway, worth a try.
This is a scene where the HRSC nadir channel is used to pan-sharpen the colourised MOC data (corresponding to the MOC+HRSC images in my previous post):
This is the same scene, now using the nadir channel to pan-sharpen the actual HRSC colour channels, before low frequency-matching them to the MOC colour reference:
The HRSC colour data definitely make the scene more dramatic (and hopefully provide higher frequency colour information), not to mention replacing fake colour data with real ones (at least at high spatial freqencies).
I noticed some slight striping artefacts in the HRSC colour channels, but hopefully they won't be apparent close up in Orbiter. At the lower resolution levels I might gradually fade out the HRSC colours.
I've got one more idea up my sleeve I want to try before starting on the global mosaic in earnest: Currently I am matching the low resolution colour reference by rescaling the low freqency band of the RGB channels to match the means of the reference. While this matches the mean intensity, it doesn't match the contrast. This can be a problem if the brightness difference between HRSC fragment and reference data is high. For example, if the HRSC sample is much darker than the reference, rescaling the means might lead to a much higher contrast than the reference.
To address this, I am going to try to match the low-frequency variance as well as the mean. This would match variance multiplicatively, and mean additively. Of course, this operation might subdue the drama of the above scene, so I'll have to see how it looks. Anyway, worth a try.