uncertainty in georeferencing

Phil Crossley

I am georeferencing 4 different images (from 4 dif. decades) to a contemporary orthophoto.

My understanding is that the accuracy of each rectification depends partly on how many control points I use, where they are within the image, and how precisely I can select the same place on both the unrectified and the initial orthophoto.

Given these factors (and in my case, it is not easy to find the same number of control points with the same distribution or confidence of precision), I am wondering if others have tested, investigated, or thought about how confidently I/we can later claim that the coordinates of any given point on an image will be the same as the coordinates of that same point on one of the other images? In my testing so far, the "same point" has coordinates (UTM, WGS 84 in all cases) that differ by 3-10 meters.

Have others found this? What kind of uncertainty have you reported for your subsequent analysis using a series of georeferenced images?

thanks in advance,
Phil Crossley

global_mapper

Phil,

The accuracy of the rectification will depend on several factors, including your accuracy of placing your control points and also the projection that you use for the rectification. In theory if you rectify in the original projection of the image, just 2 control points can be used for a perfect rectification (assuming no distortion in what is being rectified). If you are not sure of the original projection for the image, then you will likely need more control points, especially if your map covers a large area of the earth. In these cases typically 4-5 or sometimes more control points are needed. The absolute accuracy will really depend on the resolution of the image that you are rectifying and of course your placement accuracy for the control points.

Let me know if I can be of further assistance.

Thanks,

Mike
Global Mapper Support
support@globalmapper.com

Frank2

Another important factor in rectification accuracy is elevation changes. Ideally, you want to find points on level ground to avoid radial displacement. Look at the tree image at the bottom of this link. The tree is planted at the same point, but the top leans out in different directions.
http://hosting.soonet.ca/eliris/remotesensing/bl130lec4.html

Frank2

Ah, now I found the image I was really looking for. Notice how all the skyscrapers lean away from the center of the photo. Terrain elevations cause the same distortion in rectification.
http://www.ecophotoexplorers.com/images/wtc_aerial_lg.jpg

Phil Crossley

Mike. Thanks. I'd like to explore this further.
I am using aerial photos of different scales and resolutions taken over a 40 year period, and comparing each to a recent digital orthophoto that is in UTM coordinates based on WGS 84 (actually, on ITRF 92, but its very close to WGS 84).

I'm assuming there IS distortion in each of the photos, of the types mentioned by Frank2 and others, and I know that I cannot select exactly the same point in both my unrectified and my orthohoto images--though I'm sure trying. So, I'm assuming there is uncertainty.

Because of this, I'm using 20 control points, and trying to spread them ~evenly across my image, but only within the region that I know to be relatively flat (and I'm not planning to do any mapping or measuring of the areas 'outside' this area). My original image thus has no projection (or are you referring to my orthophoto?) except the 'central perspective' inherent to all vertical aerial photos.

But that still leaves me with the questions: how much variability should I expect in the UTM coordinates reported for any randomly selected, same point, between the georeferenced images created for each decade. I can prove that there is variation, just by putting my cursor on the same canal bank, for example, in the georeferenced image from 69, then in the one from 83, and the original orthophoto--I'm just wondering what kind of variability others have found, how you documented it, and what you would consider acceptable (when subsequently measuring areas on each of the images for example), or, whether you have found such uncertainty to decrease with one of the georeferencing methods vs. another???

thanks for any suggestions,
Phil

Frank2

I haven't done any rectification myself in years. I think the last time I actually did it was in the 90's when I wrote my own code to do so. Typically you just try to get the RMS within some allowed tolerance, and either you or the program discards the control points with the worst errors. The accepted US government photos seem to allow a wide margin: "The horizontal accuracy of these DOQs followed the National Map Accuracy Specifications that 90% of the points on image are within 40 feet of their true position."
http://74.125.95.132/search?q=cache:rpdE1kwx-wQJ:www.wou.edu/las/physci/taylor/luck/landuse/image_registration_RMS_error.doc+rms+aerial+rectification&cd=2&hl=en&ct=clnk&gl=us

Phil Crossley

Thanks for the link to that document, and for that great WTC aerial photo in your earlier post which I'll use in my classes in the future.

pimskoy

Hello all,

I never use Global Mapper for rectifying image before. Could I do X, Y, and Z value rectification in GM?
I have quickbird image that I'd like to rectify to a lidar dem.
Thank you in advance.

Putri.

global_mapper

Putri,

Global Mapper currently only supports XY rectification. Orthorectification is not supported.

Thanks,

Mike
Global Mapper Support
support@globalmapper.com