[gdal-dev] Thin Plate Spline

[Jan Hartmann]

Thanks Peter, good points. One further question: even if the 
georeferenced map as a whole is distorted due to too few control points, 
the control points *themselves* should always be projected to their 
exact specified georeferenced location, right? At least that is wthat I 
thought was the difference between rubber sheeting algorithms like thin 
plate spline and polynomial rectifications, where some kind of error is 
distributed between the control points.

Oh, and I would be very much interested to know more of your experiences 
with the algorithms to be used for rectifying distorted historical maps.

Regards,

Jan

On 05/04/10 08:16, Peter J Halls wrote:
> Jan,
>
>    the number of points required will depend upon the nature of the 
> deformation in the originals.  You need sufficient control points to 
> ensure that the algorithm can sort things out, so relatively more 
> points are required in the deformed areas.  However, although I've not 
> used the GDAL Thin Plate Spline, with many of these algorithms it is 
> necessary to also have an even spread of control points, or new 
> distortion can be introduced in the previously undistorted areas.
>
>    I would regard five control points as the minimum and suitable only 
> for originals that were completely flat when scanned.  Five points 
> should be enough (just) to deal with orientation errors on the scanner 
> and systematic scanner errors.  If the original had been folded, for 
> example, even had it been ironed before scanning, additional points 
> would be needed either side of the former folds, in order to sort out 
> the compression they introduce.  The worst case I've ever had to deal 
> with was a late 19th century map with no known projection details that 
> had been segmented and mounted on linen - with gaps between each 
> segment.  In that case, I ended up transforming each segment and then 
> stitching the segments together - it worked remarkably well.
>
> Hope that helps,
> best wishes,
>
> Peter
>
> Jan Hartmann wrote:
>> Hi,
>>
>> I'm rubber-sheeting old, deformed maps using the thin plate spline 
>> option of gdalwarp (-tps). My impression was that the control points 
>> would be transformed to exactly their georeferenced locations, but in 
>> the georeferenced image the control poimts don't quite align with the 
>> gcp-coordinates I specified. I'm only using five points per map, so 
>> could this be the problem? If so, how many points do I need to get 
>> the control points on their exact location? Or are there other ways 
>> to achieve this?
>>
>> Jan
>> _______________________________________________
>> gdal-dev mailing list
>> gdal-dev at lists.osgeo.org
>> http://lists.osgeo.org/mailman/listinfo/gdal-dev
>