Hello Erik, Am Mittwoch, 25. Juli 2018 17:15:19 UTC+2 schrieb Erik Krause: > > Am 25.07.2018 um 16:38 schrieb klaus...@gmail.com <javascript:>: > > Please compute, in cartesian coordinates, the partial derivative > d^2/dxdx > > of the 2-dim mapping function with c=1 as parameter. > > Hint: you'll encounter some x/abs(x) - like terms. > > > > > > > >> In the wiki discussion you write "sqrt() has a singularity at 0". This > >> is not true. sqrt(0) is 0 > >> > > In simple language: singularities can be present even if the function is > > continuous. > > And how would this affect the image? For no correction at all the > function is linear with a slope of 45°. If there are a, b and c > parameters the d parameter is adjusted such that the curve always goes > through (1,1). For some parameter values it is not 45° at (0,0), but > this results in a different magnification in the center, like to be > expected if you want to correct pincushion or barrel distortion. >
The main point is that when you overlay remapped images, they will not register correctly. There are areas that work well, and other areas where image content is offset. If you do a hard blend at the line segment bisector (Mittelsenkrechte), lines across it will not have a step but will have a kink. My math is too rusty to argue with you, but I would be interested > whether you can estimate how large the error would be? That depends on the lens, and also on the zoom setting. My observation is that for one of my old cameras when zooming the a and c parameters changed sign. I also observed that in such a case using only the barrel b parameter gave CP errors a few 1/10th pixels in size. What intrigued me, after optimisation with b, then switching on a and c, alignment for CPs near the image center improved, but a good (sharp, well-defined) CP near the image edge or corner worsened in alignment from 3 to 5 pixels for instance. > Will it be as > large or larger than the error f.e. introduced by parallax due to > entrance pupil shift (which is pronounced for fisheye lenses but also > present for rectilinear wide angles). > That depends on object distances. You tell me real world values, but let's assume 1 cm entrance pupil shift. For an object in 10 m distance that is 1 mrad. Now I assume a lens with one rad or 57.3 degrees opening angle. Say 4000 pixels hence a shift of 4 pixels. A fisheye is more like 3 radians. For simplicity 3000 pixels. Shift of 1 pixel. Adjust numbers to your situation. Parallax due to entrance pupil shift, you can compensate to some extend in repositioning your camera on your nodal point adapter. Parallax errors due to objects at different distances, the seem finder in enblend does a quite good job in putting seems away from the problem zones. Best regards Klaus -- 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/e48b5eca-ddd3-44ba-aa21-d49372139b89%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
