On Sunday, 29 September 2019 22:58:09 UTC+1, lukas wrote: > > > you mean lower frequency, right? >
No, higher (I said highEST but I meant highER). Lower frequencies would be expected to be used, because they will have spread from valid areas. You don't want to see the higher frequency components though (which will include the hard edge in Enblend's case, although I'm not 100% sure on how it works). > Do you know how many pixels are maximally included to both sides (by > means of diffuser layers) in enblend and multiblend? I have never > noticed an artefact where there is a grey fringe without black which > should result from the seam being close to the edge but not beyond the > edge. > That depends on the number of levels, which varies. In Enblend's case I think it's related to the size of overlap as each image is added (so some images may get blended more or less than others). In Multiblend, it depends on the size of the images (either input images or output image, depending if --wideblend is specified). For the same number of levels, Multiblend blends about half as much as Enblend does, but it probably picks a higher number of levels. I think the total "spread" of a pixel's influence varies with the number of level cubed. NFT is of course very close to the edge at the two entry points and if > either the coarse mask or the optimisation mess up, it might end up on > the wrong side ... > > Is there an easy description of how they are constructed? Do you > explicitly penalise close proximity to the edges? > To clarify, Multiblend's seams are still precise in a general sense, they just operate with a different metric. Enblend calculates the true Euclidean distances, which I think is one reason it is slow (it's optimised, but the algorithm is still complicated), whereas Multiblend uses a propagating algorithm, increasing distance by (if I remember rightly) 3 for every horizontal/vertical pixel and 4 for every diagonal pixel (only addition and comparison required). That results in a distance function which is more of a fat octagon than a circle, but for calculating seam lines it makes no practical difference to how "good" they are, unless you've somehow managed to take geometrically perfect photos, which never happens. So there's no explicit penalty for being close the edges, but I'd say at worst Multiblend might stray about 5% closer (in a Euclidean sense) to the edges than Enblend. -- A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ --- You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group. To unsubscribe from this group and stop receiving emails from it, send an email to hugin-ptx+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/3d86da2b-4556-44f7-9b60-bbb3b8495b93%40googlegroups.com.