GEOM30009 IMAGING THE ENVIRONMENT

Assignment 4
Image Georeferencing and Orthorectification


Due for submission at 10:00 pm on Friday of Week 12


Objective
Value: 10% of Subject Mark
The aim of this assignment is to learn how to georeference a satellite image, generate a Digital
Elevation Model (DEM) from a pair of images and use the DEM to orthorectify an image. You will
use a stereo pair of IKONOS imagery taken from Mount Wellington near Hobart, Tasmania. You will
learn how to rectify the images using a set of Ground Control Points (GCPs). Further, you will
generate a DEM from the stereo imagery. Then, you will orthorectify the images and assess the
correctness of orthorectification by overlaying a vector map of the region on the generated
orthophoto.

Background
Georeferencing and orthorectification are critical steps to enable geometric measurements on
remotely-sensed images. Georeferencing refers to the relationship between the image coordinate
system and a geographic coordinate system. The pixels in a non-georeferenced image do not have
ground coordinates (e.g., X, Y, and Z) which are fundamental for measurements such as distance
and area. During the georeferencing process, the effect of image perspective (tilt) will be removed.
On the other hand, orthorectification is the process of removing not only tilt distortion but also
relief displacement. This enables overlaying existing vector maps on the image with good accuracy.
Given the image coordinate and the geographic coordinate of a set of GCPs, e.g. road junctions or
building corners, the georeferencing can be done by estimating a transformation between the
image coordinate system and the ground coordinate system. Orthorectification is performed by
using a DEM and a set of GCPs. The georeferenced and orthorectified image can be used in many
applications when metric information is crucial.

Data
A pair of satellite imagery, along with a parcel map of Hobart will be used for this assignment. The
dataset can be downloaded from LMS under “Assignment4: Image georeferencing and
orthorectification”.

Software
ENVI 5.6 will be used for reading the dataset and GCPs, estimating the transformation and
generating a georeferenced image. To generate a DEM, a pair of images with sufficient overlap is
used. Following the DEM generation, an orthophoto will be generated for either of stereo images.

Tasks
The assignment consists of three tasks:
1. Georeferencing a satellite image and its quantitative accuracy assessment.
2. Generating a DEM using a stereo pair of imagery.
3. Generating an ortho image and verifying the orthorectified image using a fine-scale map. The
whole assignment should be completed within three weeks.

Task 1: Georeferencing a satellite image
In the task, you will georeference a given satellite image using a number of GCPs with known ground
coordinates provided in a pts file.

Steps:
1. Copy the required files from the LMS to your local disc area. Extract the folder isprs-
ikonos-dataset-hobart.rar.
2. Start ENVI 5.6. and open the IKONOS image set: File > Open As > Optical Sensors >
DigitalGlobe > IKONOS
3. In IKONOS window, open the “isprs-ikonos-dataset-hobart” and choose the text file
“po_162779_metadata.txt” in the folder “po_162779_0000000”. Repeat this step for image
“po_162780_metadata.txt”.
4. From the toolbox (on the right side of the software), click Geometric Correction and then click
Registration and select Registration: Image to Map.
5. For either of images, choose the R (i.e., band 1), G (i.e., band 2) and B (i.e., band 3) bands
according to RGB layers of the image and press OK.
6. In the Image to Map Registration window, set the projection parameters as UTM (project
system), WGS-84 (Datum), Meters (unit), 55 (Zone), S (South), 1 meter pixel size and press OK.
7. From the menu in the Ground Control Point Selection window, select File > Restore GCPs
from ASCII and choose the right corresponding GCP file *.pts for the selected image. For
example, if you work on po_162779 image, you should select the GCP file
GCPs_georef_162779.pts.
8. Having selected the GCPs, the software allows you to see the calculated RMS error for all
GCPs in the lower left corner of the Ground Control Point Selection window. However, this
value is beyond the acceptable range, meaning the transformation model is not
accurately estimated.
9. To accurately estimate the transformation model, the polynomial degree should be changed
to 3 or 4. To do this, you can simply increase the degree in the Ground Control Point
Selection window. Also, by clicking on Show List, deactivate those GCPs whose RMSE is too
high. This can be done by simply clicking on on/off icon in Image to Map GCP List. An RMSE
of 3 pixels or smaller is considered acceptable for this task. To easier exclude GCPs with
large error, you can go to Options and sort the GCPs according to their error.
10. Once the georeferencing parameters are estimated, in the Ground Control Point Selection
window click Options and then select Warp Displayed Band. Generate a warped image and
the image will be added to the Layer Manager window. Remember not to close the Ground
Control Point Selection window, since you will need it in the next steps.
11. Now every pixel in the image should have ground coordinates. To see the ground
coordinates of a pixel, you can click Cursor Value icon indicated by or simply press Ctrl + I.
12. To assess the accuracy of the georeferenced image, choose 10 ground control points that
have not been employed in the georeferencing process and are distributed across the image.
Determine their ground coordinate (Cursor value) from the georeferenced image, then
compute the errors (Equations provided below). These 10 points are check points since they
have no role in georeferencing estimation. In order to find the location of each check point on
the georeferenced image precisely, first zoom in to the image (e.g., to 60%) and find its map
coordinate (i.e., Map X, Map Y) from the list of points (in the Ground Control Point Selection
window). In Go To box (next to drop down list Annotation on the tab bar), insert the Map
coordinate (Map X, Map Y) to moves to the view screen, which has the proximity of the
wanted point. From here, navigate to the corresponding point of the check point on the
georeferenced image to measure its coordinates with the cursor. You can find a guide
describing locations of points on the image, together with their ground coordinates in the
GCP’s-Hobart.doc file.

Note: In your report, you should discuss how the generated georeferenced image is different from the
original image, and calculate the RMSE over the check points by computing the difference between
their real ground coordinate (in the GCP’s-Hobart.doc) and the corresponding ground coordinates
(cursor values) measured on the georeferenced image.

To calculate the Root Mean Square Error (RMSE), first calculate the error of X and Y coordinates for each
point i using the following equation:



(1)

Consequently, the mean error in X and Y (MEX and MEY) is using the following equations:



(2)

Finally, calculate the Root Mean Square Error in X and Y (RMSEX and RMSEY):



(3)

and the Combined RMSE (RMSEP):

→

(4)
In the above equations, the superscript chk refers to coordinates of check points determined from
the georeferenced image (cursor value) and ref indicates the real ground coordinates of checkpoint
from the given text file. You can save the corresponding values in an Excel file for further calculation.
Steps for error calculation using the 10 check points selected in the previous step:
1. Identify and fill into an Excel sheet the X and Y coordinates of the provided check points in the
georeferenced image (Xchk, Ychk) through right clicking on the layer and choosing Cell
Coordinate.
2. Calculate error in X and Y per point (i.e., and ) using equation (1).
3. Calculate the Mean Error in X and Y (i.e., MEX and MEY) using equation (2).
4. Calculate Root Mean Square Error in X and Y (i.e., RMSEX and RMSEY) using equation (3).
5. Finally, calculate the combined RMSE (i.e., RMSEP) using equation (4).
6. Save the Excel file that includes the derived X, Y coordinates and the results of calculations.

Note: In your report, you should discuss the factors affecting the RMSE of the georeferencing
process.

Task 2: Generating DEM from a stereo imagery
We will generate a DEM of the region which will be used for orthorectification.

Steps:
1. Open ENVI.
2. From toolbox, find Terrain- DEM Extraction and double click on it.
3. Select DEM Extraction Wizard: New.
4. In the opened window click Select Stereo Images and from the pop-up window click the drop-
down button Open, select New File, then navigate through the image files and select tif format
of each image, i.e. po_162779_rgb_0000000.tif and po_162780_rgb_0000000.tif
5. Choose po_162779 band R as the left image and po_162780 band R as the right image. Click
OK and go to the next step by clicking on Next in DEM Extraction Wizard window.
6. Select radio button Read GCPs From File and select GCPs_Hobart-2.pts provided and then by
pressing OK proceed to the next step.
7. Choose Read Tie Points From File, then select the 26_18-1.pts.
8. In step 6, change the directory of outputs as you wish otherwise leave it unchanged and go to
the next step.
9. In step 7, be sure that parameters are correct, especially x and y pixel size must be set as 1
meter.
10. In step 8, choose the directory for storing the DEM file.
11. In step 9 by clicking Load DEM Result To Display, you enable to see the result of generated
DEM and visually evaluate it. Click finish and do not save the current state of the wizard.
12. In order to see the generated model in true color, first you should close all the DEM
windows, then click the 3D SurfaceView from the Toolbox and in the opened window, choose
the RGB layers of either of the images and press OK. In the next window select the DEM file
and click OK. Now in the next window, choose the parameters that you would like the 3D
model is created based on. Select DEM resolution to 512 and the image resolution to 1024.
This setting is dependent on the memory of the system you are using and in case of errors,
reduce the resolutions. Other parameters could be left as their default. By pressing OK you
would see the 3D model of the generated DEM.

In your report, you should include a screenshot of the generated DEM and provide a visual
interpretation of the elevations and the topography of the area.

Task 3: Orthorectification
Having generated a DEM, we can orthorectify the image to remove the effect of relief
displacement.

Steps:
1. Open ENVI.
2. Navigate through toolbox Geometric Correction> Orthorectification> RPC Orthorectification
Workflow.
3. In the opened window, choose your input file which can be either of stereo pair image and for
DEM file select the generated DEM file.
4. In the next step, choose the GCPs_Hobart-2 file. You see that all the GCPs will be listed and
there is a green check next to each point indicating the contribution of points to
georectification.
5. Since the horizontal accuracy is quite high and it must be in a range of a few meters (less than
3 meters), some GCPs should be removed. To do so and be able to see the maximum error, the
button shows error overlay must be on. Further, for GCPs that have an error greater than 3
times of their standard deviation, there is an inclination mark next to the point. These points
should be deleted by clicking the icon with a red cross . Also, by visually checking which GCP
has significant error, delete the point. Repeat this step until sufficient accuracy is obtained.
6. To evaluate the accuracy of image orthorectification, we should have some independent
points which don’t contribute to the process. To have these points, change the status of 10
points from Adjustment to Independent. Try to select these points across the image so that
they cover the entire image.
7. Now by going to Statistics you enable to see horizontal and vertical accuracies for adjustment
GCPs as well as independent GCPs. Summarize the result of accuracies in a table and bring it in
your report in a very clear way.
8. Once the orthophoto has been generated, load the shape file of parcels from folder
LIST_PARCELS_HOBARTS, i.e. list_parcels_hobart.shp. This vector map should precisely
overlap with the orthophoto image. Take a snapshot of both layer and include it in your
report. You should include the entire data as well as a few zoomed-in snap shots verifying
both layers overlap properly.


Submission
Write a 1500 ± 10% word (at max 2500) scientific report and include the following content:
1. Provide a proper introduction with background information, addressing the purpose of
georeferencing, DEM generation and orthorectification of satellite imagery.
2. In the Methods section describe the process you performed to complete each of the three
tasks. You can also include intermediate results if they help you describe the process.
3. Include the result of each process in result section. (e.g. snapshot of georeferencing result, 3D
model of the DEM, and the orthorectified image overlapped with the vector map, accuracy
results for georeferencing and orthorectification.)
4. In the Discussion section provide an analysis of your results and address the following
questions:
i. How does the distribution of GCPs impact on the accuracy of georeferencing? What is the
role of GCPs in georeferencing?
ii. As noted, the key difference between a georeferenced image and an orthorectified image
is the removal of relief displacement. Describe what this effect is and how it is removed
during the orthorectification? You can associate your answer with some illustrations.
iii. Assuming an accuracy of 15 m for the DEM and a field of view of 1 degree for the IKONOS
image, what is the maximum radial error of the orthorectified image?
5. Provide a clear and concise conclusion summarizing your findings.
6. Provide a reference list for your supporting research.


Submit a digital version of your report via LMS and in pdf format only.
Marking rubric
Appropriate length and proper formatting 5%
Proper introduction 10%
Proper method section 10%
The georeferenced image present and correct 10%
The RMSE correctly calculated and reasonable related to georeferencing 10%
The DEM image present and correct 10%
The orthophoto overlaid with vector map present and correct 10%
The RMSE correctly calculated and reasonable related to orthorectification 10%
Three questions answered and properly discussed 15%
Logical conclusions 10%