Georeferencing - Paul Bolstad at Work

advertisement
Image Georeferencing
Lesson 4: Image Georeferencing
What You’ll Learn: We introduce image georeferencing, also known as image
registration or image transformation. We’ll then update a data layers through
vector digitizing.
You should read the section “Coordinate Transformation”, in Chapter 4 of the
GIS Fundamentals textbook before starting this Lab.
Data: The several data layers are in the \L4 subdirectory:
Scand.jp2 is an unregistered, scanned image in arbitrary scanner coordinates,
with a 0.9 meter cell size.
Big_Mar_Roads, a road data layer for the study area, in NAD83 UTM zone 15
coordinates, units are meters
BigMar2008.img, a registered image from 2008 in Washington County
Coordinate system, HARN NAD83, units are survey feet.
BigMarE1938.img, BigMarN1938.img and BigMarSW1938.img, unregistered air
photos from 1938.
Background: Image data are a common source of information, and particularly
useful when mapping vegetation, wetlands, and land use change. While many
images are provided in digital formats and already “registered” to projected
coordinate systems, many times they are not. This is particularly true for
historical images, which often are only available in paper or film media. These
are often scanned, and must then be georeferenced. Georeferencing an image,
often called an image transformation, converts the image from a file or scanner
coordinate system to a projected map coordinate system.
There are many forms of image registration, but the simplest is called a firstorder or affine transformation. An affine transformation is appropriate when the
terrain is flat and the photograph has been taken with a vertically oriented
mapping camera. When the camera is not pointed straight down (vertical), then
at least a projective transformation should be used, and if there is significant
surface height variation, then a complete geo-correction, or photo resection,
should be performed. In this lab we’ll just cover the affine transformation.
Control and Image Data
Image georeferencing require both a source for geographic coordinates (our
control layer), and a target layer to be transformed. In our case the target layers
are images, but there are times when we wish to register point, line, or polygon
feature layers. Here, our targets are aerial photographs in an arbitrary scanner
coordinate system. This system has been slightly modified to make viewing
easier, but is representative of what most scanners produce.
1
Image Georeferencing
The control layer, in this case is a vector roads layer, is the source for your
control information. This control layer must be in the desired map coordinate
system. You will identify points in the control layer, and match them to
corresponding points in the target image. When you have enough points and are
satisfied with them, you create the transformed image (Video: Georeference).

Start ArcGIS - ArcMap, and open a blank map.

Add Big_Mar_Roads.shp. This is our control layer, the one from which we
will get control locations. You need to load (open) this data layer first.

Add scand.jp2, our target layer. Ignore the ‘No Coordinate System’
message, as the point of georeferencing is to provide a coordinate system.

Save the project use a sensible name and place it with the \L4 data.

Notice how the two data layers are in different locations on view. If you can’t
see both layers, click the zoom to full extent button,
and you should
now see both the image layer and roads layer (see figure below)

Target Layer
Control Layer
2
Image Georeferencing

Make the Big_Mar_Roads.shp layer active (left click on it) then right
click and select Zoom to Layer. This will put the roads into the center of the
display. We need to see all the roads to begin Georeferencing.
Look for the georeferencing toolbar, shown at the bottom of the figure below. If
this toolbar is not displayed, right click on any toolbar, and make sure there is a
checkbox next to the Georeferencing entry.
Inspect the Georeferencing toolbar, and make sure the Layer: is scand.jp2. This
is the target layer
Note the toolbar buttons
View Control Points Table
Select Control Points
Rotate the image
Left click the triangle to the right of the
Georeferencing drop down box, and
Left click on Fit to Display
3
Image Georeferencing
This will bring the target and control layers to be roughly coincident (see below).
First, we’ll attempt to manually rotate and move the image fit the roads.
Use the Rotate cursor or the Shift cursor to try and move the picture to match
the roads. You will notice you can’t do it. Sometimes you get close with one
road, but at the expense of mis-alignment with others. That is because the roads
layer and the unregistered image are not only out of alignment but also differ in
scale.
To clear you manual adjustments errors, delete scan.jp2 and add it back again,
repeating the zoom to roads layer and fit to display commands.
4
Image Georeferencing
Georeferencing an image is done using the Control Point Cursor.
Left click on the icon to activate the cursor.
Now left click to mark a succession of control links, or points, between the target
layer and the control layer.
Note that the selections should always proceed as:
 First select a point in the target layer (the image),

then select the corresponding point in the control layer (the roads in
this case)
Zoom in on the view to a portion with a road intersection. Zoom in enough so
that you may clearly see the width of the road, and easily position the cursor in
the middle of the road
on the image (please
see the figure).
First, carefully place
the cursor over the
Carefully place the
cursor in the center of a image of the road
intersection and left
road intersection in the
click
image, and left-click.
Now, carefully place the
cursor over the
corresponding road
intersection in the
vector layer, and click.
If all went well, you
should see the image
shift slightly to match
the feature a bit more.
Second, place the
cursor over the feature
road intersection and left
click
It is best to pick well
distributed points, so for
your second point
select an intersection
on the right side of the image.
Repeat this procedure, panning and zooming to a different road intersection.
First, click on the image version (target layer), then on the feature layer version
(control layer).
If you lose your link line after you zoom or pan, click once on the Control Point
button. This should display your active line.
5
Image Georeferencing
With each successive control link, the image should line up a bit better.
After your 4th link examine the Control Link Table (Control Link Table button)
Save Link Table
when finished.
This is only for documentation;
you cannot recreate this process
from a saved link table
Notice that after the 4th link you get a report of the current point error, and the
RMS error in the lower right Link Table. This reports an index of the accuracy of
your transformation. Strive to keep your mean number below 13, and if you can,
below 10.
Continue until you have at least eight control links. If you make a mistake, open
the Link Table, select the erroneous row and push the Delete key.
One tip: You can often select more than eight points and delete the ones that
contribute the least to the fitting equation. The Residual numbers on the right
side of the Link Table tell which ones are most helpful, those with the smallest
residuals. Remember to keep your points well distributed, don’t delete all the
high residuals so that your retained points are all on one side of the image
Write down your final Total RMSE value. You will later add this to your map.
6
Image Georeferencing
When you are done, save your points table from within the Control Link Table
(save button, shown on graphic above), carefully noting the file name and
location (something like BigMar_tab2000.
From the Georeferencing Toolbar select Rectify to save the image to a file,
Browse to your jump drive Output Location, select IMAGINE Image format and
name the output something like BigMar2000.
Ensure the file
name is has the
extension .img
Remove you unregistered photo (scand.jp2) from the Map and add your new
registered image (BigMar2000.img).
Now digitize the current boundary of Big Marine Lake on your rectified image:
 From ArcCatalog, Create a new Shapefile, name it Lake_2000, specify as
polygon, and import the projection from the other layers in the map.
 Add this new empty lakes layer to your map
 Make sure the editor toolbar is activated, set the lakes layer as your target,
and digitize the boundary of Big Marine Lake as it existed in 2000. Use a
scale of about 1:2500. (see the key at the end of this exercise to understand which is
Big Marine Lake)

When creating large complex polygons it may be useful to digitize the lake a
section at a time. This can be done by carefully tracing the adjoining lines
using the Trace cursor and merging (Editor Dropdown) the adjacent polygons.

Another alternative is to create a section of the lake, leave it selected (blue
color outline), then use the Reshape feature tool to “add on to the 1st section.
7
Image Georeferencing

You are not required to “cut” out the islands inside the lake. It you wish to do
so, after you have completed the lake, make Lake_2000 “hollow” and create a
new polygon for a island, then select the Editor menu and select “clip”,
discarding the area that intersects; finally delete the island polygon.
We’ll now register a historical aerial photograph, taken near the height of the dust
bowl in the 1930s. We’ll then digitize the lake boundary then, and compare it to
the current lake boundary.
Registering Historical Aerial Photographs
You will now complete three more registrations.
First, create a new data frame (Insert – Data Frame), and add the image file
BigMar2008.img. This is a bit lower resolution, and a bit larger area than the
image you used above. Look at the coordinate system, and units. This image is
in the Washington County Coordinate System, HARN NAD83, in U.S. feet,
verifying this by looking at the data set properties.
Load the image you just registered (named something like BigMar2000.img)
place it on top of the BigMar2008 image. Remember, it is doing “on the fly”
reprojection, so the images should line up on top of each other, or nearly so. If
not, check the properties for the image you rectified (right click on the name in
the TOC, then Properties). Does it have the Coordinate system correctly listed
as a UTM NAD Zone 15N, in meters?). If not, fix it (ArcToolbox – Data Management – Projections
and Transformations – Define Projection).
After you have the two images displayed correctly, click the top image on and off,
zooming to inspect the edge, to see how well your registration turned out.
You may also make the
top image transparent,
and compare the outline of
the lake to the image
below it in the view.
To change the
transparency, right click on
the top image, then
Properties – Display, and
set the Transparency to
50% (demonstrated for
BigMar2000.img, but the
process should work for all
images).
8
Image Georeferencing
Now remove the BigMar2000.img from your data frame
Create two (2) new empty data frames. Add BigMar2008.img to each of them.
The select the first data fame with BigMar2008.img and “activate” that layer. (Hint:
Activate is right click on the data frame title and select Activate.)
Next add BigmarE1938.img to this data frame. Right click on the data frame title
and in Properties, rename the data frame to Big Marine 1938 East.
Repeat this process to add the BigmarN1938.img to the second data frame with
BigMar2008.img in it. Rename the data frame to Big Marine 1938 North.
Repeat the process again add the BigmarSW1938.img to the third data frame
with BigMar2008.img in it. Rename the data frame to Big Marine 1938 South
West.
Your screen should now look similar to what is shown below:
.
Activate the East 1938 data frame. This contains an historical black and white
aerial photograph, taken during the dust bowl years, and perhaps the earliest
photo with a metric camera of the area. This image and two more historical
photos you will process have been scanned, but not rectified to a coordinate
system
Register this bigmarE1938.img image to the Washington County coordinate
system, U.S. Feet, using the BigMar2008.img as a source for control points.
9
Image Georeferencing
Remember to:
FIRST, carefully study the 1938 image. Many features have changed in 72
years. Get oriented with the image.
Then zoom to the approximate location of the 1938 image on the BigMar2008
image.
Use “Fit to display” to drape the 1938 image over the BigMar2008 image.
Next select the control points first in the target (unregistered 1938 image), and
then select the corresponding location in the control (color BigMar2008 image)
As so many features have changed over the years you will be able to only get 4
or so points. Make sure they are as well distributed as possible.
See suggested control points below: (note the red o’s)
You may get strange flips and distortions when selecting control points because
the error of point selection dominates, and leads to transformation equations that
are wildly off the mark (and you may also get these strange flips after the first
four points, but it is much less likely). If it does occur, it is best to delete the 1938
image, and sometimes the whole data frame, and start again.
Use the process described in the previous section to register, and then rectify the
image. On saving the image, name it something like 1938East.img, so that you
can identify it later.
10
Image Georeferencing
Activate the Big Marine 1938 North data frame and register that image following
the previous instructions. (See 1938 East)
See suggested control points below: (note the red o’s)
Name the output 1938North.img
Activate the Big Marine 1938 South West data frame and register that image
following the previous instructions. (See 1938 East)
See suggested control points below: (note the red o’s)
11
Image Georeferencing
Finally create a new data frame and load all three registered images. They
should look something like the figure to the right:
Use ArcCatalog to create a polygon layer named Lake_1938. Import the
coordinate system from the BigMar2008.img.
Now, digitize the lake boundary as it appeared in 1938. You may want to make
adjacent images somewhat transparent to move from one image to another.
The boundaries for a greatly reduced Big Marine Lake are apparent by the black
areas in the photos. Note that in some photos the lake water is a bit lighter than
others, due to sun/camera angles, so look at all three photos in common areas to
best identify the lake boundary. Look at the example below for guidance on the
approximate boundary.
12
Image Georeferencing
Map
Composition:
Compose a map
similar to the figure at
right, containing the
transformed image
(BigMar2000) from the
first part of the
exercise, the roads, the
lake boundary in 2000,
the lake boundary in
1938, a scale bar,
north arrow, title, and a
legend..
Make sure you include
the RMSE of your first
transformation.
Remember, you are
combining data from
two different
projections, the image
you rectified (UTM
NAD83 zone 15N) and
a Washington County
Coordinate System,
HARN NAD83, in U.S.
feet, so you have to
make sure the
projections are properly assigned in each data layer, and select the appropriate
datum transformation (think about the two projections you are combining when
selecting among the datum transformation options).
13
Download