Loading ...
Sorry, an error occurred while loading the content.

Re: d e parameters

Expand Messages
  • Erik Krause
    ... always and individual for each image. Stitching of scanned images would be much worse without. ... Not only for rectilinear lenses, for all kinds in fact.
    Message 1 of 7 , Oct 2, 2006
    • 0 Attachment
      On Monday, October 02, 2006 at 10:44, JD Smith wrote:

      > How commonly do people find they need to allow non-zero d and e parameters

      always and individual for each image. Stitching of scanned images
      would be much worse without.

      > (which change the center of the radial aberration coefficients a, b, and c)
      > for rectilinear lenses?

      Not only for rectilinear lenses, for all kinds in fact. And not only
      for the center of a, b and c correction but for the complete lens
      remapping. However, they are not so much needed for longer focal
      length lenses...

      > How large of a shift is typical? Does anyone have
      > recommendations for good test images to assess whether d and e should be
      > nonzero (standard "building at a distance"?).

      I simply optimize them and the result is much better afterwards. For
      a digital camera with a fixed focal length lens optimization of per
      image d and e parameters might not be necessary. For our D70 they are
      pretty constant at about 13 and 9 pixels...

      best regards
      --
      Erik Krause
      Resources, not only for panorama creation:
      http://www.erik-krause.de/
    • JD Smith
      ... Sorry, yes I meant for non-scanned digital images. I find about 20 and 18 pixels on my D50, but it likely depends on the lens. If you allow per-image d
      Message 2 of 7 , Oct 2, 2006
      • 0 Attachment
        On Mon, 02 Oct 2006 20:48:37 +0200, Erik Krause wrote:

        >> How large of a shift is typical? Does anyone have recommendations for
        >> good test images to assess whether d and e should be nonzero (standard
        >> "building at a distance"?).
        >
        > I simply optimize them and the result is much better afterwards. For a
        > digital camera with a fixed focal length lens optimization of per image d
        > and e parameters might not be necessary. For our D70 they are pretty
        > constant at about 13 and 9 pixels...

        Sorry, yes I meant for non-scanned digital images. I find about 20
        and 18 pixels on my D50, but it likely depends on the lens. If you allow
        per-image d and e on your D70, how much variation do you find? I could
        see that being a somewhat poorly constrained optimization direction.

        JD
      • Sacha Griffin
        You might confuse semi-advanced people with always and individual for each image. For film, yes, digital no, unless, you re using a zoom lens. For example
        Message 3 of 7 , Oct 2, 2006
        • 0 Attachment
          You might confuse semi-advanced people with " always and individual for each
          image."
          For film, yes, digital no, unless, you're using a zoom lens. For example my
          28-135 has got loads of SLOP, (D & E Shift)

          "I simply optimize them and the result is much better afterwards. For
          a digital camera with a fixed focal length lens optimization of per
          image d and e parameters might not be necessary. For our D70 they are
          pretty constant at about 13 and 9 pixels..."

          However, remember everyone just because you can get your average pixel error
          down, doesn't mean your stitch is better.
          Optimization is an imaginary set of distortion points supposedly matched to
          a set of images. The "error" is the difference between your points and your
          optimization parameters combined with your image locations. You can get a
          zero error by hitting every imaginary point, but not matching your real
          images. The best stitch finds a happy medium with a spread of control
          points, and low error.

          Sacha Griffin
          Southern Digital Solutions LLC
          www.southern-digital.com
          www.seeit360.net
          www.ezphotosafe.com
          404-551-4275
          404-731-7798


          -----Original Message-----
          From: Erik Krause [mailto:erik.krause@...]
          Sent: Monday, October 02, 2006 2:49 PM
          To: PanoToolsNG@yahoogroups.com
          Subject: [PanoToolsNG] Re: d e parameters

          On Monday, October 02, 2006 at 10:44, JD Smith wrote:

          > How commonly do people find they need to allow non-zero d and e parameters

          always and individual for each image. Stitching of scanned images
          would be much worse without.

          > (which change the center of the radial aberration coefficients a, b, and
          c)
          > for rectilinear lenses?

          Not only for rectilinear lenses, for all kinds in fact. And not only
          for the center of a, b and c correction but for the complete lens
          remapping. However, they are not so much needed for longer focal
          length lenses...

          > How large of a shift is typical? Does anyone have
          > recommendations for good test images to assess whether d and e should be
          > nonzero (standard "building at a distance"?).

          I simply optimize them and the result is much better afterwards. For
          a digital camera with a fixed focal length lens optimization of per
          image d and e parameters might not be necessary. For our D70 they are
          pretty constant at about 13 and 9 pixels...

          best regards
          --
          Erik Krause
          Resources, not only for panorama creation:
          http://www.erik-krause.de/




          --


          Yahoo! Groups Links
        • Erik Krause
          ... I mean the stitch of course, not the average error (speaking of scanned images). The mentioned d and e parameters for the D70 where the result when
          Message 4 of 7 , Oct 2, 2006
          • 0 Attachment
            On Monday, October 02, 2006 at 15:14, Sacha Griffin wrote:

            > "I simply optimize them and the result is much better afterwards. For
            > a digital camera with a fixed focal length lens optimization of per
            > image d and e parameters might not be necessary. For our D70 they are
            > pretty constant at about 13 and 9 pixels..."
            >
            > However, remember everyone just because you can get your average pixel error
            > down, doesn't mean your stitch is better. Optimization is an imaginary set of
            > distortion points supposedly matched to a set of images.

            I mean the stitch of course, not the average error (speaking of
            scanned images). The mentioned d and e parameters for the D70 where
            the result when creating a batch stitching template in a room paved
            with post-it stickers. 1514 control points where evenly distributed
            throughout all overlaps and of course I looked here more for the
            stitching and less for the pixel error (although the pixel error *is*
            a measure if you have enough well distributed points and you
            repeatedly optimize and use APCLean)...

            best regards


            --
            Erik Krause
            Resources, not only for panorama creation:
            http://www.erik-krause.de/
          Your message has been successfully submitted and would be delivered to recipients shortly.