ChameleonScales schrieb am Sonntag, 31. Januar 2021 um 19:15:50 UTC+1:
> calculation of the correlation is expensive > > I assumed the correlation was also calculated when you detect CPs. That > said I didn't necessarily mean to show it at all times. Just when you press > Fine-tune all would be enough, but still in a separate column. > But then the column would be the most time 0, then there will be other complains. > Cpfind searches in different levels, correlation is only calculating in > 100 % scale. > > Do you mean if you stitch photos with different resolutions, CPfind with > find good CPs but the correlation calculation will give low values? Or does > it also have implications on photos with the same resolution? > Cpfind is working internally on different levels, I did not meant images with different resolutions. So cpfind can also find good cp at a higher level - which can have low correlation values (which is only calculated at 100 % scale, but the cp is evaluated by cpfind at another level ). > To explain why I imagine correlation next to distance being useful: > I assemble natural landscape panoramas which have to be stitched as > precisely as possible. > Sometimes the wind moves leaves in a turbulent way and trees 200 meters > from the camera may look slightly different between 2 photos. > On the other hand, if there's a stone 3 meters from the camera, the > correlation and number of CPs is going to be higher on it because the wind > doesn't affect it. Problem is: if the tripod is not properly calibrated to > avoid parallax (I'm not always the one taking the photos), the far away > trees may be more reliable even though they have low correlation, but Hugin > may prefer the stone because there are more CPs on it. This means the > optimizer will give a higher CP distance on the trees which will introduce > a seam in the far away landscape. > "You should delete the CPs on the stone" I hear you say. The thing is I > don't always know if there is parallax and sometimes, close elements take > up such a large part of the photo that I'd prefer not to have CPs covering > a small area, so I leave the ones on the stone in the hopes that there's no > parallax. > > And here's the point: > If I could see that the CPs with low correlation have a high distance, > this would draw my attention and I may find that these CPs are in the far > away trees and then I would know not to trust that stone and delete those > CPs with more confidence. > The correlation is only calculated at a small region (default 21x21 px). So if there is a parallax error this is not necessarily visible in the correlation value. And cpfind is not using this correlation value - it uses another algorithm. So the correlation value say nothing about the quality of a cp. (The correlation is only used by the fine-tune algorithm, nothing else uses this value.) And a single cp has (in most cases) no big effect on the whole pano. All cp work together. So if you have cp on the "stone" and the "tree" the optimiser will try to satisfy the alignment of both elements. But if you prefer the alignment in one of the elements leave the cp only on this element and remove the cp on the other. But this is a matter of the image content and of personal taste which element is more important, so there is no general algorithm how to solve this problem. -- 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 [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/afab9b2c-3cfb-4880-8fb8-9921853d5a47n%40googlegroups.com.
