Weird Hugin's exposure/vignetting correction

[Max Lytvynov]

Dear community.
I have faced strange behavior of Hugin's exposure and vingetting correction.
I am stitching Mars Perservance' panoramas from original NASA raws published here https://mars.nasa.gov/mars2020/multimedia/raw-images/

Filter: SOL 2, Camera: Mastcam Z (left)

This is a 7-row 360 degree pano consisting of 142 shots at 100mm camera of rover

First I started stitching few weeks ago and I faced improper exposure normalization which was producing results like this for the first 2 rows:

as well as for the entire pano:

Despite various exposure anchor images that I used, this just was making this sharp edge be in diferent parts of panorama.

Using some manual adjustments and custom  exposure correction tricks I got more or less acceptable result, but vignetting correction was still far from good

After I finished entire pano I found out that due to high parallax of rover on foreground there are missalign glitches even on parts where only Mars surface exposed, and I also found out that Hugin allows to select only certain photos to be bend and aligned instead of rebuilding whole pano.

So i decided to start from the beginning by stitching surface only parts firt and then bend and align rover parts but keeping surface unchanged.

For this time Hugin threw me another surprise and now this produces some weird vignetting correction for the same photos that I used last time but treats exposure properly...

There are not too much options that affects photometric alignment so idk how to affect hugin's behavior on this point.

Any suggestions ?

[Bob Mahar]

Hi,

For photometric optimization to model vignetting there needs to be significant overlap between different annular regions of the images.   Regular / disciplined shooting patterns can produce a situation where the edges of the images match up exposure wise.   So the difference in exposure between the the two images at the control points is not significant.   Even so, it cannot model vignetting if no central portion of an image overlaps - essentially the edges are corrected at the expense of the center.   I think that's what you are seeing.  There are a couple ways around this: 

(1) Find an image pair with significant overlap between different annular zones, a 50% overlap is ideal.  You only need one!  Take a suitable pair of ~50% overlapped images into a project, detect control points and add a bunch evenly covering the overlap area, do geometric alignment, and then do photometric alignment.   Hugin will be forced to reconcile the radially varying exposure between the images.   The resulting lens and camera response values can then be  applied to all images in your actual project.  Then do a custom photometric alignment disabling ( unchecking ) the vignetting parameters.

(2) Work out the vignetting curve yourself.   Examine the man page for fulla - which comes with Hugin.    It describes their mathematical model for vignetting.   its possible to play with the parameters passed on the fulla command line and have it produce various versions of a test image, if you find ones that work, plop them into Hugin.   Its also possible to put it into a spreadsheet and what-if.   Also "fulla is your friend" for batch correction.

(3) Steal the curve from other software.   If you can use another software to correct the vignetting to your liking, ( GIMP, RAW Therapeeeee, etc. ) for a single image where their vignetting parameters are visible.   Do a little math and you have values to use in Hugin.   The models likely won't be similar, but once you have the response curve from one, you can model that in the other.

(4) See if NASA publishes the vignetting parameters or if they have a lenstip profile.    You never know and googling is cheap.   Obviously they have to correct these images too, there may be research papers which leak the values.   I know they have used Hugin.

-- Bob