|
|
GEO 4120c Workshop 16 March 2009 |
Dear 2009 Students: Please
note that I wrote the incorrect file names for the two images in the
second part of this exercise (image-to-image registration). The new
file names, which are the same two images (although the reference image
is a copy of the earlier one) that are used in the first part of the
exercise, are now included in the instructions. I'm sorry about any
inconveniences, and will not require that the exercise be turned in at
the usual time.
MWB 10:00 PM, 29 March 2009
Introduction
This week you will go through an ERDAS Imagine exercise to learn how to correct the geometric errors inherent in aerial photographs. The tour guides teach you which buttons to push, but not why. This workshop is set up to show you how to teach yourself why you are inputting certain information.
In this workshop, you will:
Scanned images may
have features such as roads and rivers
that may allow you to associate the data with a geographic location.
However,
raw data do not represent locations on the surface of the Earth unless
the data
carry some reference to the ground location. Geometric rectification
(or
georeferencing) is the process by which the geometry of an image area
is made
planimetric, thus allowing the data to be used to provide accurate map
locations for features in the data. The process almost always involves
relating
GCPs (ground control points e.g., meters in northing and easting for
the
Transverse Mercator map projection) to pixel coordinates from the image
(e.g.,
row and column values). This is necessary whenever accurate area,
direction,
and distance measurements are required. Most serious earth science
remote
sensing research is based on analysis of data that have been correctly
rectified to a base map.
Workshop
Method
I - Image to Map Rectification with a Digital Map
Originally, Image to Map
rectification was done by measuring the X and
Y, latitude and longitude, or easting and northing coordinates of
points on a
paper map, usually a topographic quadrangle (often 1:50,000, 1:100,000,
and
1:250,000 scales) and then typing in the coordinates as the references
(see
below). These coordinates were then associated with a point on the
image
designated with the GCP Editor (see
below), and the interpolation models were calculated. Many areas of the
world
will still require this approach because the original maps are not
available
digitally. Recently, georectified, digital images of many of the
developed
world’s topographic quadrangles and rectified high spatial resolution
aerial
photography is available for registering both aerial photography and
satellite data. This exercise, while nominally
an image-to-map rectification, uses a properly registered digital
orthophoto quadrangle
(DOQ) for the reference points in the same way that an image-to-image
rectification would occur. The difference here is that the reference
data are
in a sufficiently fine spatial resolution.
Two basic
operations must be performed in order to geometrically
rectify a remotely sensed image to a map coordinate system:
1. Spatial Interpolation, which
defines the
nature of the geometric coordinate transformation that must be applied
to
rectify or relocate every pixel in the original input image to its
proper
position in the rectified output image.
2. Intensity Interpolation, which
is the mechanism for
determining the brightness value to be assigned to each pixel in the
rectified
output image.
Imagine offers
several methods to rectify images to maps. In this
exercise you will perform a simple rectification of an unrectified
Landsat 7
ETM+ image to a UTM map projection. The procedure follows the general
outline:
1. The GCP Editor is used to
create .gcc
files containing image (row and column values) that relate to map
coordinates
(UTM meters) for selected ground control points.
2. The .gcc file is input to the Transformation
Editor, which
creates a matrix containing the transformation coefficients.
3. A geometric model file (.gms) is then
created using the model
properties and the original .img image files are input into the Resample
Display program to produce a rectified .img image file.
You are now going
to use a Digital Orthophoto Quadrangle (DOQ) of
The scanned image
TIFF
file G:/share/GIS
4120c Aerial Photo
Interpretation/Air_Photo_Example_Data/gainesville_1979_NHAP_600dpi.tif
should be copied over to your own
folder.
Open the MrSID file gville_doq.sid
and the TIFF file gainesville_1979_NHAP_600dpi.tif
into separate viewers. Place both images in the viewers before you
begin
rectification.
Go to your viewer displaying the TIFF image
of
Gainesville and find and select the Geometric Correction... button
in
the Raster dropdown menu. A menu should appear that allows you
to select
the Geometric Model. Select Polynomial and click OK. The
Polynomial
Model Properties window should appear as well as the Geo
Correction
Tools. We are going to be using a Polynomial Order of 1 (default).
Now
let's start the GCP Editor by clicking on the crosshair button in the Geo
Correction Tools box
Once
you click on this button, the GCP Tool Reference Setup menu
should appear. Select Existing Viewer, select the viewer with
the DOQ
and click OK. This should bring up a dialog box asking you to
select a Reference
Layer. Go to your directory and click on gville_doq.sid and
click OK.
When the Reference Map Information box appears click OK
again
and wait for all windows to position themselves. The GCP Tool window
should appear on the bottom of the screen and two Chip Extraction
Viewers
(magnifiers) should appear, one for each viewer. These viewers are to
assist in
the placement of your GCPs.
If
there are already GCP point data entered, highlight all of the points
by left-clicking in the boxes in the Points# column and dragging down.
Then right-click and 'delete selection.' Now you are ready to proceed.
In
the GCP Tool window select File - Save Input As... to name a
new GCP file. Move the cursor to the blank window at the top of this
menu and
type the name ex6input.gcc (again making sure you are in your
directory). Once you move the cursor out of that window the .gcc file
extension
is automatically added. Select the OK button. Now do the same
for the
reference points you will add. Go to File - Save Reference As... and
name the file ex6ref.gcc (in your own home directory). Throughout
the GCP placement process, you should periodically save both the input
(source)
and reference (destination) files by going under File - Save Input and
File
- Save Reference. This will update your GCP locations in your .gcc
file. By
saving both the input and reference GCPs will allow you to load them at
a later
time using the Load options under the File pull-down menu.
Now
examine the DOQ a little closer. You should be able to recognize
the similar road networks, golf courses, athletic stadiums, and other
features
within your image. You will now use this DOQ to rectify your image. Imagine also supports a robust vector module
that allows a variety of editing functions found under the Vector menu
in the viewer. If you are rectifying from a vector file, all the
attributes
associated with the opened vector layer can be viewed under Vector
-
Attributes. You can change the viewing properties (i.e. color,
style) under
Vector - Viewing Properties. You may want to save your viewing
properties you have assigned as a symbology file (.evs). By doing this,
you
don't have to reassign the colors every time you reopen the vector
file. Now go
to the viewer displaying the DOQ file and place the cursor inside the
viewer.
Notice that the UTM coordinates appear in the lower left-hand corner of
the
viewer. You will now select Ground Control Points (GCPs) using UTM
coordinates
for image rectification. GCPs consist of two pairs of x, y coordinates:
- source
or input coordinates, which
are usually data file coordinates in the image being rectified, and
- destination
or reference coordinates, the
coordinates of the map or reference image to which the source image is
being rectified.
When
selecting GCPs, collect points evenly distributed throughout the
entire area to be georeferenced. This will aid in a good rectification.
Features like road intersections, corners of large building complexes,
and
land/water points etc. are good choices for finding locations on both
images
and topographic maps. A good number of GCPs for this exercise is 15-20
(in the
real world the number could reach 100’s). Now go to the GCP Tool window
and select View - Tools... The following box should appear:
The
crosshair button is for dropping a GCP onto either the source or
destination viewer. If you had no reference map or image, you would
have to
enter the coordinates by hand or enter a GCP file collected with a
Global
Positioning System (GPS). Now position the chip extraction viewer to a
corresponding
area on both the image and the road coverage. Drop GCP #1 by clicking
on the
crosshair button and then moving to the Landsat image and clicking on
the
location (street intersection) you want to place the GCP. Now repeat
this
process, but this time drop a GCP in the corresponding point on the
DOQ.
Remember that the road coverage represents street centerlines so
placement of
GCPs should take this into consideration. When you are finished with
the
placement of both the source and destination locations for GCP #1, you
should
be sent to the GCP#2 row automatically. Note: you can fine-tune the
location of
your GCPs by selecting them with your cursor and using the arrow keys
to move
them to their desired position. Also, you may want to change the color
of your
GCPs by going to the GCP Tool window and placing the cursor over the
Color row
of the corresponding GCP. Hold down the right mouse button over the
color and
select a new color.
After
you have placed 3-4 GCPs, you may want to use the GCP Prediction
feature by going to Edit - Point Prediction.
Reference from Input will predict reference points from
input points
and Input from Reference will predict input points from
reference
points. The GCP Matching feature allows you to compare
spectral values
when registering an input image to a reference image (raster to
raster).
During
the GCP selection process, you may wish to use the Automatic
Transformation Calculation button (see menu above). This will display
the
coefficients of the transformation matrix and calculate the RMS error
between
the file points and map points. For this exercise you will try to get
the Root
Mean Square (RMS) error to be less than 15.0. In the GCP Tool window
use the
scroll bar at the bottom to move your view to the far right-hand side.
Here are
the listings of the errors (residuals, RMS error and contribution for
each
GCP). This can help you determine the GCPs you might wish to change. If
the RMS
Error for all your GCPs is below 10, then congratulations, you are
talented at
selecting GCPs. The Total RMS Error for all GCPs is listed just right
of the
buttons in the GCP Tool (for example: (Total) 6.4179)). Try to get the
total as
low as possible. If the Total RMS Error is above 15.0 m, you will need
to
either delete a GCP to improve the RMS error or move the GCPs until the
RMS
error figure is more appropriate. You might want to see the RMS error
be
automatically calculated by grabbing one of the GCPs with your cursor
and
dragging it around the image, but be sure to put it back in its proper
place!
The
procedure for deleting a GCP from the table is to move the cursor
to the far left-hand side of the table and in the first box of the row
relating
to that GCP you select the lmb (the whole row should turn yellow) then
select
the rmb and in the quickview menu that appears find and select the Delete
Selection option. Try not to delete too many GCPs - it's better to
try and
make them fit before removing them.
Resampling
and Rectification
Once your RMS error is below 15.0 m you are
ready to
resample the image. Click on the Display Model Properties button
in the
Geo Correction Tools window. Browse through the Parameters,
Transformation, and
Projection folders to view the various settings. Now save this
geometric model
by clicking on Save and naming the file ex6.gms. Now close the
Model
Properties window and click on the Display Resample Image Dialog button
in the Geo Correction Tools window.
 |
Display
Model Properties |
|
|
Display
Resample Image Dialog |
When
the Resample window opens, find the Output Cell Sizes boxes.
Change both the X and Y to 30. (This indicates the size of the pixels
for the
rectified image; 30 x 30 m for ETM+ Data) Select the nearest neighbor
resampling algorithm (it should already be selected). Now name the
output image
ex6.img (in your directory), click OK, and let the resample
model do its
stuff. Your rectified image will be created and stored under the
filename which
you have designated, (ex6.img). Save it as a .tif file as well. Close
down the GCP Tool window by going under File
- Close and saving all files. Now start a new viewer to display
ex6.img and
compare the unrectified image to the newly rectified image. When
finished, make
sure you have the ex6.img image in your home directory.
Method
II - Image to Image Registration
Now we will perform an image to image
registration,
we will use the same two images. Copy
from the Share folder to your own folder the file G:/share/GIS
4120c Aerial Photo
Interpretation/Air_Photo_Example_Data/gainesville_1979_NHAP_600dpi_ex2.tif.
Bring up the gainesville_1979_NHAP_600dpi_ex2.tif file in a viewer and go to Raster ->
Geometric
Correction. Then follow all the instructions listed in the Method I,
but when
the GCP Tool Reference Setup menu appears, click on Image
Layer (New
Viewer). Your reference image is again the gville_doq.sid file.
Using
image to image registration techniques, you will locate reference
points on
your SPOT scene and locate the same corresponding points on the
unrectified TM
scene. When finished collecting GCPs and you have an RMSE near 1,
resample the
TM image using nearest neighbor with an output pixel size of 30 x 30
meters.
Then evaluate your accuracy by overlaying your TM scene onto the SPOT
scene and
using the Swipe Tool under Utilities.
Answer the following questions. Add a screencapture of the unregistered and the rectified images to the document, labeling each one and then hand in the document by e-mailing it to mbinford@geog.ufl.edu, or by placing it in your G:\share\GEO4120c Air Photo Interpretation\yourfolder\ and e-mail me to tell me that you've completed the assignment.
1)
Suppose you selected a number of GCPs that were taken from both
natural (fork in the river) and man-made (road intersection) features.
Briefly
summarize the reasons why some of your GCP's might have higher or lower
initial
RMS errors than the others. How might the attributes of images like
study area
location and resolutions of the sensor system used (e.g. spatial,
spectral
etc.) lead to easier or more difficult rectifications of the images.
2)
Note the dataset you are using for ground truth (e.g. A Digital Line
Graph or a previously rectified image). What kind of error could be
associated
with this approach to GCP acquisition?
3)
What other sources could you consider using for obtaining ground
truth for GCP acquisition?
4) In
the exercise, erroneous ground control points are deleted to
improve the geometric fit. Sometimes however, it may be necessary to
raise the
RMS tolerance level (i.e. not enough points). What would this mean in
terms of
the reliability of the rectification?
5) Is
the spatial location of your GCPs (i.e., their distribution
across the image)(in reference to scene rectification) important to
ensure an
acceptable rectification? Why or why not?
6)
What's your definition of the ideal Ground Control Point?
