[Geotiff] ModelPixelScaleTag
Pascal Peuch
peuchpascal at wanadoo.fr
Mon Sep 6 16:28:16 PDT 2004
Hello Ed,
You tend to make reading softwares as resilient as possible and I understand your point of view. Nonetheles, we have to access this issue with strict application of geotiff specification unless we have defined specific interchange conventions between a producer and a reader out of the scope of general purpose Geotiff specification.
The geotiff specification states that :
"However, tiepoints are only to be considered exact at the points specified; thus defining such a set of bounding tiepoints does not imply that the model space locations of the interior of the image may be exactly computed by a linear interpolation of these tiepoints."
and
"For orthorectification or mosaicking applications a large number of tiepoints may be specified on a mesh over the raster image. However, the definition of associated grid interpolation methods is not in the scope of the current GeoTIFF spec."
When providing three tiepoints, even if exists an affine relation within these three tiepoints, the producer doesn't tell the user that the raster space and the model space are affine related. He only tells him the position of three points.
Knowing and exchanging the earth position of one or more points of the raster is one thing but knowing and exchanging the relation between a raster and the earth is another thing.
Geotiff provides three tags to define the relation between model space and raster space : tiepoints, pixelscale and Modeltransformation. With these three means and the set of constraints of Geotiff specification, you can define the three following types of images :
1/ relation between raster space and model space is a known and simple mathematic relation :
1A this relation is X =ax+b, Y=a'y+b' (where x and y are raster coordinates and X,Y are model coordinates). In other words, the image is rectified, it has been resampled according to a projection system.
1B this relation is X =ax+by+c, Y =a'x+b'y+c' and can't be reduced to 1A case.
2/ The relation is not known or is too complex to be defined with Géotiff means.
If you are a producer :
The 1A/ case must be defined with One TiePoint and the pixelscale. (This is the normal case for corrected satellite images for example).
The 1B/ case must be defined with Modeltransformationtag (the case is not recommended by the spec and in my opinion the case is not really usefull).
In the 2/ case, you only can provide a certain number of TiePoints. (this is the case of an unrectified aerial photo or a scanned paper map without any geometric correction for example)
If you are a reader :
1/ If you find a tiepoint and a pixelscale, you know that the image is 1A type. If there is more than one tiepoint, you can check the consistency of the exceeding tiepoints and warn the user if not.
2/ if you find three tiepoints : you can check whether these 3 points are related by an affine relation. If they are you can make the assumption that the relation is this same affine relation as well for all the other points of the raster. But it is an ASSUMPTION. I understand that the users want their software to read every Géotiff file they can encounter but you just can't make such assumptions in any kinds of software (military applications for example) without asking the user at least.
See also the note in chapter "3.2.1. Unrectified Aerial photo, known tiepoints, in degrees."
Remark: Since we have not specified the ModelPixelScaleTag, clients reading this GeoTIFF file are not permitted to infer that there is a simple linear relationship between the raster data and the geographic model coordinate space. The only points that are know to be exact are the ones specified in the tiepoint tag.
----- Original Message -----
From: "Grissom, Ed" <ed.grissom at intergraph.com>
To: "Pascal Peuch" <peuchpascal at wanadoo.fr>; <geotiff at remotesensing.org>
Sent: Friday, September 03, 2004 3:25 PM
Subject: RE: [Geotiff] ModelPixelScaleTag
> In a previous message, Pascal Peuch said:
> > I think that a reader should not consider that a Geotiff file is
> > georeferenced when only a list of tiepoints is provided [...]
>
> Agreed, to a point. I think that you are pointing out some of the pitfalls
> that users or software developers might fall into, and I agree with all of
> your comments on that. However, the tie-point case is no less a valid
> method of georeferencing than either of the other two.
>
> Bottom Line, readers must support the list of tie points and, given only 3
> points, an affine relationship pretty much has to be inferred. Warnings
> might be a good idea, but are not required by GeoTIFF. With 4 or a small
> number more points, the affine relationship should be the first assumption,
> but it can easily be tested and if the relationship is not affine then the
> app needs to handle that case (including warnings, or asking the user, or
> saying it does not support non-affine or something). For a really large
> number of points, a mesh might be the best first assumption.
>
>
>
> --
> ed grissom
> ed.grissom at intergraph.com
>
>
>
> > -----Original Message-----
> > From: Pascal Peuch [mailto:peuchpascal at wanadoo.fr]
> > Sent: Friday, December 31, 1999 5:40 PM
> > To: Grissom, Ed; Gillian Walter; geotiff at remotesensing.org
> > Subject: Re: [Geotiff] ModelPixelScaleTag
> >
> >
> > I think that a reader should not consider that a Geotiff file is
> > georeferenced when only a list of tiepoints is provided unless it
> > has some external information on the geometry of the image. When
> > the Geotiff file contains only 3 (or 4 or 5 or more) tiepoints, you
> > are not granted that the file is rectified (orthogonal in the
> > coordinate system). You know the coordinates of 3 (or 4 or 5 or
> > more) points of the image and you know nothing on the other
> > points. The image could be of very raw geometry. A raw aerial photo
> > with 3 or 4 tiepoints is not affine related for example. In such
> > cases the reader software can nonetheless georeference the image
> > (depending on the accuracy requirements of the application) but it
> > should warn clearly the user.
> >
> > On the other side, producers should not use the three tiepoints
> > method to specify an affine relation.
> >
> > Pascal Peuch
> >
> > ----- Original Message -----
> > From: "Grissom, Ed" <ed.grissom at intergraph.com>
> > To: "Gillian Walter" <gillian.walter at atlantis-scientific.com>;
> > <geotiff at remotesensing.org>
> > Sent: Thursday, August 26, 2004 3:59 PM
> > Subject: RE: [Geotiff] ModelPixelScaleTag
> >
> >
> > >
> > > There are three ways to specify the raster-to-model relationship
> > >
> > > 1) point and scale (no rotation, raster is orthogonal in
> > > coordinate system)
> > > 2) list of tie points (at least three points for affine
> > > relationship, unlimited for "mesh")
> > > 3) transform matrix.
> > >
> > > I would venture to say that specifying _only_ the tie points
> > > REQUIRES 3 or more points. I would not consider a file
> > > georefernced with less than that (absent other info).
> > >
> > > --
> > > ed grissom
> > > ed.grissom at intergraph.com
> > >
> > >
> > >
> > > > -----Original Message-----
> > > > From: geotiff-bounces at remotesensing.org
> > > > [mailto:geotiff-bounces at remotesensing.org] On Behalf Of Gillian Walter
> > > > Sent: Thursday, August 26, 2004 9:24 AM
> > > > To: geotiff at remotesensing.org
> > > > Subject: [Geotiff] ModelPixelScaleTag
> > > >
> > > >
> > > > Hi,
> > > >
> > > > I have a question about the Geotiff ModelPixelScaleTag. On page
> > > > 25 of the Geotiff specification, there is a diagram indicating
> > > > that either ModelPixelScaleTag and ModelTiepointTag, or
> > > > ModelTransformationTag alone will be used to geocode an image.
> > > > However, the next page indicates that ModelPixelScaleTag and
> > > > ModelTransformationTag are optional. Is it valid to specify
> > > > only the ModelTiepointTag and not ModelPixelScaleTag? I was
> > > > under the impression that it was, since the ModelPixelScaleTag
> > > > and ModelTransformationTag parameters can't always accurately
> > > > represent a dataset's geocoding (eg. slant range SAR imagery),
> > > > but one of our customers disagrees, and I can see how the spec
> > > > could be interpreted either way.
> > > >
> > > > Gillian
> > > >
> > > > _______________________________________________
> > > > Geotiff mailing list
> > > > Geotiff at remotesensing.org
> > > > http://xserve.flids.com/mailman/listinfo/geotiff
> > > >
> > > _______________________________________________
> > > Geotiff mailing list
> > > Geotiff at remotesensing.org
> > > http://xserve.flids.com/mailman/listinfo/geotiff
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/geotiff/attachments/20040907/61858c2f/attachment.html>
More information about the Geotiff
mailing list