Hmm yes. Honoring a mask/alpha channel seems a very reasonable way to make this work.
On Sunday, February 13, 2011 4:25:22 PM UTC+1, kfj wrote: > > > It seems wasteful, but you also have to consider the time you have to > invest to pre-position the images roughly so the ROI can be detected > as roughly that area where the pre-aligned images overlap - and the > time it takes to extract/mask the ROI. Scanning the whole image > assures that the images can be positioned and oriented in any way and > all overlaps will be found. > > There is another aspect which becomes more important with fisheyes and > stereographic images: if you feed the CPG with a partial image, it > can't decipher the warp to adapt it's feature point detector: the > image geometry is quite different from center to margin. So you can't > just use partial images, but you have to use the whole images and mask > them appropriately. Masking means introduction of additional data into > the process, either in the shape of a separate mask or as an alpha > channel. Using an alpha channel might be a reasonalbly inexpensive and > transparent (hah) way of doing the needful, but I'm not sure if the > current CPGs honour alpha channels - I rather doubt it. > > Finally, if you use a reasonable overlap of 30% and calculate how much > of your images is left over in areas which are not overlapping, you > may find out that these areas are quite small after all, since the > overlap is on all margins. Even with 25% overlap, much less than half > of the image is outside overlaps. > > If you take all of this into account, I think the gain is not worth > the effort for everyday work. On the other hand, there may be special > situations where the savings would be significant. To cater for these, > the mechanism of limiting the scan for feature points to a ROI should > be available as an optional feature. In fact, this sounds like an > ideal scenario for a plugin. I'd expect stuff like this to be among > the first things to be implemented as a plugin as soon as the plugin > facility becomes maintream (currently verification of the implemented > mechanism on Mac OS is pending, but I hope it won't be too much > longer). > > I am currently toying with this mechanism for use in another demo > plugin, but I want to throw in rewarping of the parts of the images > that correspond to the ROIs to a common projection to make them > geometrically as similar as possible, thus improving CPG performance > especially with fisheye images and avoiding warp-related CPG problems > - some sort of high-end matching which would produce very good > quality, well-distributed CPs, particularly for applications like lens > calibration. Also, it would be nice to get quick access to the warped > partial images for visual inspection. > > To put a final tag on it - nice to have, but not crucial. > > Kay -- You received this message because you are subscribed to the Google Groups "Hugin and other free panoramic software" group. A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ To post to this group, send email to hugin-ptx@googlegroups.com To unsubscribe from this group, send email to hugin-ptx+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/hugin-ptx