
org.geotools.referencing.operation.builder.package.html Maven / Gradle / Ivy
Show all versions of gt-referencing Show documentation
package org.geotools.referencing.operation.builder
A package of convenience classes which use control points common to two
data sets to derive empirically the transformation parameters needed to
convert positions between the coordinate systems of the two data sets.
Note the implementation is currently (January 2008) limited to two
dimensions. The methods, however, are generic and could be expanded to
three dimensions someday. At that time, we will probably make minor
changes to the API.
The package consists of two types of convenience classes: the various
builder classes which use a set of individual control points to obtain
the conversion and the GridToEnvelopeMapper class which derives the
conversion from a grid range to a georeferenced Envelope.
The builder classes should be used by users who have two data sets that
are known to share certain common points but who currently do not line up.
This could be the case, for example, if a user has two data sets
describing the same region but one of these has an unknown coordinate
referencing system. In this situation, there is no way to convert
coordinate positions between the two data sets. However, if the user can
identify a series of positions coupled in each data set, a Builder can
calculate an empirical conversion between the two data sets. The
different Builder classes use different mathematical approaches to obtain
the empirical estimate.
The GridToEnvelopeMapper should be used by users who have a grid, such as
an image, which is not georeferenced but the user knows the grid is
aligned in one of the four cardinal directions and the user can identify
the outer georeferenced envelope of the grid. The Mapper class can then
calculate an empirical conversion object to map positions in the image
coordinate system to georeferenced positions.
The builder classes require a matched set of known positions, one from a
"source" data set and another from a "target" data set; the builder will
then provide a structure which contains a conversion object to transform
positions from the "source" coordinate system to the "target" coordinate
system. The builders require a list of
{@linkplain org.geotools.referencing.operation.builder.MappedPosition MappedPosition}
objects which are associations of a
{@linkplain org.opengis.referencing.DirectPosition DirectPosition} in the
"source" data set with another DirectPosition in the "target" data set. The
{@linkplain org.geotools.referencing.operation.builder.MathTransformBuilder#getTransformation() getTransformation() method}
in the builder can then be used to provide a
{@linkplain org.opengis.referencing.operation.Transformation Transformation}
object from which the user can obtain the
{@linkplain org.opengis.referencing.operation.MathTransform MathTransform}
to use for conversion operations.
Different builders use different mathematical approaches for obtaining the
empirical estimate of the conversion parameters. The builders are:
- {@linkplain org.geotools.referencing.operation.builder.ProjectiveTransformBuilder ProjectiveTransformBuilder}
- {@linkplain org.geotools.referencing.operation.builder.AffineTransformBuilder AffineTransformBuilder}
- {@linkplain org.geotools.referencing.operation.builder.SimilarTransformBuilder SimilarTransformBuilder}
- {@linkplain org.geotools.referencing.operation.builder.BursaWolfTransformBuilder BursaWolfTransformBuilder}
- {@linkplain org.geotools.referencing.operation.builder.RubberSheetBuilder RubberSheetBuilder}
with the mathematical details of each estimation procedure explained in
the documentation of the builder class itself. The first four of these use
a least squares estimation method in which, if the system is
over-determined by having more than the minimum number of control points
necessary to derive the estimate, the best matching parameter estimate
will be obtained by minimising the sum of the squared distances to the
points. The RubberSheet algorithm uses a linear interpolation between the
various control points.