home

Topics Topics Edit Profile Profile Help/Instructions Help Register Register  
Search Last 1 | 3 | 7 Days Search Search Tree View Tree View  

Correction of non-regular distortions

The Panorama Factory Forum » The Panorama Factory FAQ » Correction of non-regular distortions « Previous Next »

Author Message
Top of pagePrevious messageNext messageBottom of page Link to this message

Guenter Giese
New member
Username: Guenter

Post Number: 1
Registered: 11-2003
Posted on Sunday, November 02, 2003 - 6:33 am:   Edit PostDelete Post

John Strait suggested that, before calculating panoramic views from a series of images, one should first correct for distortions (barrel or pincushion) to obtain undistorted images. But in some cases, it is impossible to obtain images without visible distortions:

Some lenses produce "non-regular" distiortions, i.e distortions which cannot be corrected using barrel or pincushion correction tools. An example is the Nikon 18-35 mmm zoom lens creating some "moustache-like" distortion (barrel distortin near the center and pincushion distortion near the edges) in a certain zoom range. Panoramic views of buildings are amongst the most critical themes.

As with all zoom lenses, the type and amount of distortion is dependent on the focal length.

Does anybody know a program to correct such distortions? The software (or plugin) ideally should be able to perform a transformation (shift) of the " real" imaged points to the "ideal" points (like warping an irregularly distorted grid to a rectangular grid).

The ultimate solution would be able to do this transformation automatically by first calculating, from a set of imaged templates (e.g. real images of rectangular arrays of points or grids), the "transformation functions". In a second step, these (zoom-lengh dependent) correction parameters would be used to correct the real, distorted image.

This sounds straightforward, but I guess it is very hard to achieve (no simple mathematical function may be available).

Guenter
Top of pagePrevious messageNext messageBottom of page Link to this message

John Strait
Moderator
Username: Jstrait

Post Number: 25
Registered: 5-2001
Posted on Monday, November 03, 2003 - 3:37 pm:   Edit PostDelete Post

Guenter Giese's suggestion is a very interesting one.

My advice to correct barrel distortion before stitching was made before the V3 release of The Panorama Factory.

A correction for barrel distortion was introduced in the V3 release of The Panorama Factory. This means that there is no longer a need to correct the barrel distortion using another application. Instead, you should enable the correction for barrel distortion on Wizard step 2.

"Non-regular" distortions like the one Guenter describes in the previous post can often be corrected by The Panorama Factory's alignment fine tuning option. Alignment fine tuning makes local adjustments to the image alignment in the overlap region in order to improve image matching.

Because this correction applies only in the overlap region, it will not correct distortions in regions that do not overlap. You may wish to increase the overlap (up to 50%) so that this correction can be applied to a larger portion of the image.

During initial development of The Panorama Factory, I did some experiments with the method Guenter described of imaging a rectangular grid to develop correction parameters. This method can be very successful, but it requires a great deal of care in photographing the grid in order to accurately characterize the lens. And there is some concern that the characterization may change with zoom setting (on a zoom lens) or with focus changes.

The method used in The Panorama Factory to correlate image data in the overlap region can deliver results that are as good (in terms of improving image alignment). The advantage is that it can be done automatically for any lens without requiring the user to characterize the lens.

By the way, I actually own a lens that exhibits this "moustache-like" distortion. This is the 24mm lens I used for many of the panoramas shown in the Pittsburgh and Twin Cities galleries. The Panorama Factory's alignment fine tuning function did an excellent job of producing good alignment in these images.



(Message edited by admin on November 03, 2003)
Top of pagePrevious messageNext messageBottom of page Link to this message

Guenter Giese
New member
Username: Guenter

Post Number: 2
Registered: 11-2003
Posted on Tuesday, November 04, 2003 - 5:38 pm:   Edit PostDelete Post

John,

than you for your detailed answer. I will try out your suggestions as soon as possible.

If one is able to correct a non-regular distortion in the process of creating a panorama: can the workflow of warping be stored in detail / recalled for use with another set / pair of images? If this were the case, would it be possible to re-apply these corrections to a single image (which would kind of represent the middle image in a set of three images)?

Such a workflow could then be established for different lenses / zoom factors using images of grids etc. and doing repetitive alignments of an uncorrected test image for calculation of the warping parameters. This would be a fine tool for general distortion correction.


Guenter


Top of pagePrevious messageNext messageBottom of page Link to this message

John Strait
Moderator
Username: Jstrait

Post Number: 27
Registered: 5-2001
Posted on Wednesday, November 05, 2003 - 9:19 am:   Edit PostDelete Post

I'm sorry, but The Panorama Factory does not provide a way to save the alignment fine tuning results and apply them to a different image. It may not be appropriate to do this.

The alignment fine tuning does not really characterize distortions introduced by the lens. Instead, it makes small adjustments to the image warping to achieve better image-to-image alignment. These adjustments may or may not be a good way to characterize the lens.

Fine tuning results are influenced by the actual image data because of certain implementation details of the software. The results are also influenced by small rotations of the camera (if any) which may be different from one image pair to the next. In practice, these factors mean that the fine tuning map is unique to each pair of images.

By the way, surprisingly, photographs of grids do not deliver good results from the automatic image-to-image alignment and from the alignment fine tuning. This seems counter-intuitive, but there is a simple explanation. Regular grids and other images with repeating patterns are difficult to align correctly because there are many possible alignments (separated by one grid unit) that are almost equally good. It is difficult for the software to identify the correct choice in this situation.

Add Your Message Here
Post:
Bold text Italics Underline Create a hyperlink Insert a clipart image

Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action:

Administration Administration Log Out Log Out   Previous Page Previous Page Next Page Next Page