![]() Image Registration
Image registration
is a geometrical transformation that creates a correspondence between image
coordinates (row and column) and a reference coordinate system. In the SPRING
this reference coordinate system is, in the worst case, the planar coordinate
system for a given cartographical projection. As any cartographical
projection has a well defined relationship with a geographical coordinate
system, it is possible to say that the registration
establish a correspondence between image coordinates and geographical
coordinates. Other common terms used
for the registration procedure are geocodification and georeferencing.
It is important, however, to make a clear distinction between registration
and geometric correction. The image geometric correction procedure
removes systematical geometrical distortions introduced in the image
acquisition step, while the registration just uses simple geometrical
transformations - usually polynomial transformations - to establish a mapping
between the image coordinates and geographical coordinates. So it is
suggested to use the registration as a technique to refine the
geometrical quality of images with system geometrical correction. The registration is a
required operation to make the integration of an image to the existing
database in a GIS. For years the remote sensing projects assume that the
images could be integrated to data extracted from existing maps or to
measurements performed directly on the terrain. The registration is also
important to combine images from the same area taken with different sensors
or to multi-temporal studies, in case the images are taken in distinct
periods of time.
Polynomial
Transformations - control points
The usage of polynomial
transformations is very common for image registration. The polynomial
transformations make the relationship between image coordinates and
coordinates in the reference system through the control points. Control
points are features subject to identification in the image and in the
reference system. Road crossings, airport runways and river junctions are
natural candidates for control points. The parameters
determination of the polynomial transformation selected is made through the
solution of an equation system. In order to build this equation system the
coordinates of the control points have to be known in the image and in the
reference system. The image coordinates (row, column) are obtained when the
user clicks on the image features. The reference coordinates are usually
obtained through reliable maps which have the corresponding features used as
control points (Table mode in the registration window). The SPRING
also accepts measurements performed directly in the field using a GPS (keyboard
mode). Existing vectorial data and georeferrenced images can also be used as
sources for referenced coordinates (Window mode). Once the n
control points are determined and the polynomial transformation is selected,
a system with 2n equations is built to solve 6, 12 or 20
parameters, depending on the polynomial degree (1o, 2o
or 3o degree). Thus, it is possible to conclude that the
minimum number of control points is 3 for the 1o degree polynomial,
6 for the 2o degree polynomial and 10 for the 3o
degree polynomial (see the equations for the 1o and 2o
degree polynomials in the figure below).
One should be aware that
the control points distribution in the area to be registered is very important,
because the polynomial transformations tend to behave adequately only inside
the region where the control points are defined. Images with system geometrical correctionThe SPRING can recognize
satellite images with system geometrical correction and treat them in a
special way during registration. It is the case for TM images from LANDSAT-5,
ETM from LANDSAT-7, CCD from CBERS-2 and from high resolution images
available in the GEOTIFF format. The system correction is based on
physical parameters usage inherent to the acquisition situation for each
image (ephemerid and platform attitude, sensor imaging system and Earth reference model). As a result an image is
obtained where the pixels already observe a relative positioning related to a
certain cartographical projection system, where, usually, a residual
translation is required because of the uncertainty in the satellite positioning.
In this kind of image the internal geometry is almost always well solved and
it does not require being modeled using polynomial transformations. A potential advantage
when treating specially images with system correction is to use fewer control
points (in this case a single point would be enough) to refine the
residual translation. Another advantage is related to the fact that the
control points do not need to be well distributed by the whole area to be
registered. When an image with system
correction is read by the IMPIMA module, it is generated a file in the GRIB
format which allows the system correction option to be enabled in the
registration step. Thus, the user can treat in a special mode an image with a
correction system (system correction button activated), but it is also
possible to treat an image in the conventional way, if desired (system correction button disabled). This functionality is present in the Importing
GRIB files interface. With the system correction option activated, the SPRING access, in the GRIB file,
the navigation equation (relationship between image coordinates and
projection coordinates) from the system correction and refines the
translations (in X and Y) of this equation while the control points are
acquired. NOTICE: It is emphasized here that
besides the fact that this option works with a single control point, it is
possible to use all the good control points acquired. CBERS-2 Images with system geometrical correction
The CBERS-2 images
selected through the INPE images catalog (www.dgi.inpe.br/CDSR) has a
system geometrical correction and then, subject to uncertainties from the
ephemerid data and attitude used in the geometrical correction process. As
the CBERS-2 images have been distributed freely to the Brazilian users community, it is suitable to emphasize the actual
state of the geometrical quality of these images and the description of the
best way to treat them through the SPRING registration module with the
purpose of eliminating the positioning error and refine the internal error. The precision in the
positioning defines how much an image with system correction is out of its
correct geographical position. The CBERS-2 images can present
positioning errors up to The internal precision establish the integration possibility from an
image with system correction to maps and other georeferrenced data. In other
words, the internal error is the residual error, the one which is not
completely removed when an image is superposed on a map, that is, a small internal
error means a good superposition. The internal error is about Resampling
by interpolation
To compute the new gray
level value in the image to be registered, the SPRING adopts as interpolation
method the techniques known as bilinear interpolation and nearest neighbor allocation. The nearest neighbor
allocation interpolator attributes a gray level value to the pixel
of the corrected image, equals the gray level value of the closest pixel to
that position. There is no alteration in the gray level value. Because of
this characteristic, it can be applied in images where there is not too much
heterogeneity in the gray level values. The bilinear interpolator computes the gray level of the pixel in the
corrected image using the 4 pixel neighbors. As a result, the
gray level of the pixel is changed by the values of its neighbors. It can be
applied in images where there is a considerable heterogeneity in the gray
level values of the pixels. Check the procedures
to make a georeferrencing (registration) of an image. The image has to be in the Grid
Binary (GRIB)
format and stored in the disk. The GRIB format is generated by the "Impima", the module for reading
images. NOTICE: The registration of images
stored in other formats, such as TIFF, GEOTIFF and RAW, requires first
a conversion to the GRIB format, which can be done using the "Impima"
module. Registration
in the SPRING
The georeferrencing can
be performed using the cartographical parameters in an active project
(Window mode). If there is no active project the control points
coordinates can be informed using the digitizing table (Table
mode) or by computer keyboard (keyboard mode). See details for
each mode: TABLE: In the table mode the user needs
to have a map (topographical chart of the same image area being considered). This
map has to be calibrated
in a digitizing table. As it is not required to activate a project (it is
just enough to have an active database), the system requests the type of
projection to be used in the registration. WINDOW: In this mode the user can use an
Information Layer in an active project. This IL can be an image that is
already georeferrenced or any other map, as a thematic IL that has some
identifiable features in the image. KEYBOARD: In the keyboard case it is not
required to have an active project (it is just enough to have an active
database). Because of this, it is required to inform using planar or
geographical coordinates, which can be directly measured in the terrain
using, for instance, a GPS. See
all the procedures to register an image.
Image
Registration and Mosaicking
When the area covered by
an image is smaller than the project area, it is required to make a
mosaicking with adjacent images. An image mosaic is
just a collage process of adjacent images to have a larger coverage of a
given area. The registration between
these two images compounding a single information layer in the mosaicking
process has to be precise enough so that in the superposed region there are
no gaps or distortions in the continuity of geographical features. The error
in pixels has to be the smallest possible. Mosaicking with
TM-Landsat images
The SPRING can make an
automatic cut in TM-Landsat images, for complete scenes (full frame), not displaced from its orbit. This cut allows the maximum
reduction of the superposed area among neighbor scenes, using the nominal
area from a WRS (Worldwide Reference System). Just click
on the Cut WRS button in the "Importing GRIB Files"
window. IMPORTANT: This resource only works for
whole scenes of the TM-Landsat.
Logical
sequence to perform a registration
Next it is presented an
example for a logical sequence that the user has to follow to georeferrence
an image using a map placed in a digitizing table. The example assumes there is a Database and a Project.
NOTE: If the user gets control points
using the Window mode, substitute step 3 by "Select an information layer as a reference in the main Window". NOTE: If the user gets control points
using the Keyboard mode, do not consider step 3.
![]() |