GsE 189 Remote Sensing: Theory and Applications

Laboratory Exercise #4
Image Rectification and Registration

Submitted by:
Jasper B. Molleno

09 September, 2018
Objectives:
- To georeferenced digital images by different methods.
- To explore and evaluate different image rectification and registration techniques

Materials:
- ENVI 5.2 Software
- Scanned 1:50000 Topographic Map of Bolinao, Pangasinan
- Google Earth Screenshot Image of Bolinao, Pangasinan (Focus on Santiago Island)

Procedures:
Image Registration Using Coordinates/GCPs from Paper Maps (Image to Map)
1. The scanned image of the topographic map of Bolinao was opened in ENVI. Information
such as the coordinate system of the map was recorded for the following steps.
2. An image sketch was drawn was drawn to show the corresponding coordinates for the
corners and tick marks in border and in the body of the map.
3. From the Toolbox, Geometric Correction? Registration? Registration: Image to
Map. Bands 1,2,3 are selected for R, G, B respectively.
4. The Image to Map Registration window appeared and necessary parameters were set.
5. The Image to Map Registration was done by inputting the corresponding geographic
coordinates of the tick marks in the ground control points selection after clicking the tick
mark whose coordinates are inputted.
6. Add Point clicked to record the points in the Image to Map GCP List. Points were
adjusted so that the RMSE will be under within the acceptable value of 0.500. Position
adjustments was kept to be only along the intersection lines of the corners/tick marks so
that distortions for the georeferencing would be minimal.
7. The GCPs were save by clicking File >Save GCPs with lat/lon… from the Ground
Control Points Selection window.
8. After that, the image was warp to georeference the image. [Memory 1] was selected as the
input warp image.
9. Parameter were set afterwards and the georeferencing of an Image to Map was completed.
Image to Image Registration
10. A screen capture of Santiago Island in Bolinao Pangasinan City was taken using Google
Earth. The image taken will be used for the following steps.
11. ENVI Classic was opened since image to image registration is simpler for the classic view
of ENVI 5.2.
12. The image file and the warped topo map were opened and were loaded in band
combination of 123 for both images
13. Map from ENVI’s toolbar was access and registration of points for image to image
georeferencing was started.
14. Unlike the previous part, a point for the image and another point for the topo map were
selected to obtain a certain GCP.
15. Things such as RMSE was maintained to be under 0.50 as well as the logical choosing
points to be paired for both images.
16. After placing 17 points (which the minimum point needed for the next steps), the
screenshot image for Santiago Island was georeferenced in a combination of different
transformation methods and resampling methods.

Answers to questions:
1. Show the map sketch with the correct coordinates. How many points did you find?
Answer:

Figure 1. Simple sketch of Bolinao topo map

A simple sketch was drawn in paint. The intersection of lines in the image shows the tick
marks inside the map. 12 points were found along the sides and 5 point were found inside
the map, for a total of 17 points. Points for the 120-00-00 E longitude along the edges of
the map were not included because it does not have tick marks for this longitude.

2. Describe the relationship between the Main, Scroll, and Zoom image displays.
Answer:
Main display – It displays the whole image itself. The smaller box indicates the area to
inspect. This display can be use to move from one specific are to another in a larger scale.
Scroll display – This is a larger view for the smaller box in the main display. It shows a
zoomed in version of the area specified in the main display. In this view, panning through
the image is possible however, the panning is only limited to the area restricted by the main
display. It also has a smaller box for a more zoomed in view.
 Zoom display – based on its name, zoom display is the most zoomed in version of the
specified area. It shows the most specific feature or pixel of the image.

3. In performing an image to map registration of a scanned map, how will you compute for
the proper pixel size (3.173 in this exercise) to be indicated in the image transformation
process?
Answer: Given 400 DPI, pixel size of 3.173 was obtained using this solution

1 𝑖𝑛𝑐ℎ 0.0254𝑐𝑚
× = 6.35 × 10−5 𝑚/𝑝𝑖𝑥𝑒𝑙
400 𝑝𝑖𝑥𝑒𝑙𝑠 1 𝑖𝑛𝑐ℎ

𝑑𝑔𝑟𝑜𝑢𝑛𝑑
𝑑𝑔𝑟𝑜𝑢𝑛𝑑𝑝𝑖𝑥𝑒𝑙 = × 6.35 × 10−5 = 3.173
𝑑𝑖𝑚𝑎𝑔𝑒

4. How many GCPs are required to be added before the RMSE is computed? Why?
Answer:
5 GCPs are required to be added. Because the default settings for the first order polynomial
warp (affine) that is set in ENVI follows the rule that the number of GCPs required is a
degree plus one squared of the polynomial transformation. (sample solution: GCP> (1+1)
^2, the minimum number of GCPs which is >4 is 5.)

5. What are the causes of uncertainties in the position of the GCPs in the image?
Answer:

Orientation of the image. The image was observed to be a little distorted that the tick
marks found on the body of the map does not exactly align with each other.

Student’s judgment. Placing of GCPs for a distorted image affects how high or low the
uncertainty for the GCPs. The judgment was based on how high the RMSE value is and
that it must be within the set acceptance value that must be less than 0.500.

Resolution. The resolution of the image makes it difficult to place the correct coordinates
at their designated tick marks. Since there is no true basis on where the actual coordinates
on the map is, deviations or errors from the actual value is created for every GCP
established.

6. Note that the total RMSE required should be less than 0.50. Why? When the total RMSE
is already acceptable, have a screenshot of the list of GCPs showing the individual and
total RMS and include it in your report. Include also the screenshot of the map showing
the position of the GCPs used.
 Figure 2: Map showing GCP in topo map

Figure 3. RMSE values of each point

Answer:

7. Paste the File Information of the georeferenced image listed in the Display Manager.
Compare it with the File Information of the raw scanned map.
Figure 4. Information inside the File Information in Figure 5. Information in Data Manager after

Data Manager georeferencing

Answer:
The difference between before and after transformation is that the transformation
parameters are now shown in the “after” image file information.
8. How many GCPs for each transformation method are sufficient to correct the geometric
distortions within the image for purposes of georeferencing? Tabulate your answers.
Explain how you computed for these optimum numbers
Answer: Using the concept for determining the minimum number of GCP points for
Transformations; GCP# minimum = (degree+1) ^2 + 1

Table 1: Minimum GCP computations

Transformation Method Computation Optimum Number of GCP
RST (1st degree) (1+1) ^2 + 1 5
nd
2 degree polynomial (2+1) ^2 +1 10
3rd degree polynomial (3+1) ^2 + 1 17

9. Provide a table showing a snapshot of all GCPs (one in topo map, one in the image) per
transformation method, identify their feature types (e.g. road intersection, water edge, etc.),
GCP number (as shown in the Image to Image GCP List). Note that some GCPs can be
used in multiple transformation methods (but may have different RMS Errors due to the
different combination of points).

Table 2: Description of GCPs

Point Topo Image Description
#
1

Road
Intersection

End of
road; Near
shore

Pier?

Island
5

Road in
Dewey
Island

Shore

Sandy area

End of
road; Near
shore

Road
10

Road
Intersection

11

Vegetation
near shore

12

Road
Intersection

13

Road
Intersection
14

Road

15

Coral Reef
Edge

16

Near the
coral reef
edge

17

Near the
shore
 Table 3: RMSE Values for the GCP used in for every Transformation Method
Transformation Method
Point # RST 2nd degree 3rd degree
Polynomial Polynomial
1 0.085029424 0.15182846 0.29385341
2 0.00000000 0.19876784 0.28954985
3 0.00000000 0.00000000 0.43563477
4 0.00000000 0.12969892 0.39600983
5 0.00000000 0.11738896 0.53696390
6 0.00000000 0.00000000 0.36732615
7 0.064957521 0.00000000 0.44652447
8 0.044747979 0.00000000 0.62116080
9 0.00000000 0.00000000 0.41335257
10 0.00000000 0.079228440 0.28754786
11 0.065532304 0.20152513 0.53246810
12 0.00000000 0.11062880 0.21958611
13 0.00000000 0.00000000 0.47033672
14 0.00000000 0.17909552 0.38046197
15 0.00071241507 0.20425614 0.15032850
16 0.00000000 0.25942667 0.21300843
17 0.00000000 0.00000000 0.59779425

10. Show the resulting Total and Individual RMSE for each transformation method by pasting
screenshots of the Ground Control Points Selection and Image to Image GCP List
windows. Maximize the window to show all rows clearly. Points turned ON or OFF for
each method should be visible from the screenshots.

Image to Image GCP List Ground Control Points Selection

From above, RMSE values for RST, 2nd degree polynomial and 3rd degree polynomial. Values
show an increasing value for RMSE the more GCPs are used.
11. Provide “before and after” figures of the raw and nine warped images. Describe what
changed in the warped image as compared to the unreferenced one. Use the annotation tool
(you may switch back to ENVI from ENVI Classic) to label the images.
 Image
Transformation RAW Transformed Changes
used
The image
is titled
clockwise.
Some part
RST – NN of the
image
disappeared
because of
the rotation.
The image
is titled
clockwise.
Some part
RST – Bilinear of the
image
disappeared
because of
the rotation.
The image
is titled
clockwise.
Some part
RST – Cubic of the
image
disappeared
because of
the rotation.
The image
shows a
formation
of a curve.
2nd deg – NN
(a parabolic
curve
opening
downward)
 The image
shows a
formation
2nd deg – of a curve.
Bilinear (a parabolic
curve
opening
downward)
The image
shows a
formation
2nd deg – of a curve.
Cubic (a parabolic
curve
opening
downward)

Great
distortion
3rd deg – NN on the sides
of the
image.

Great
distortion
3rd deg –
on the sides
Bilinear
of the
image.

Great
distortion
3rd deg – Cubic on the sides
of the
image.

12. What are the noticeable differences among the nine warped images? Show screenshots of
these differences and label them. What combination gave the most accurate result? Why
do you think so? Cite your reference/s.
 3rd degree polynomial 2nd degree polynomial RST

Highest distortion for 3rd degree since it uses all the 17 point to transform the map. Higher
degree of polynomial uses more complex formulas compared to the other two method.
Meanwhile, 2nd degree polynomial shows little distortion to the ends of the map. It somehow
forms a parabolic arc opening downward. RST shows rotation for the whole map. Transition
and scale change were not that noticeable.

Conclusion:
Georeferencing in ENVI makes use of different resampling method, transformation
methods and some registration techniques. Establishing the GCPs are the most crucial part of
georeferencing since these points will be the basis for image or map transformation. Moreover,
there are a lot of factors to be considered when georeferencing such as the resolution and the
judgment of the user. After taking to account the factors that might affect the establishment of
GCPs, the users must set a tolerable value for RMSE. This value will help in acquiring minimum
errors in georeferencing.
Concepts/Tools learned from the exercise:
- More knowledge in georeferencing
- ENVI Classic (64-bit) software
References:
Enr. Claridades, A.R.C., M.Sc. “Digital image processing”. Lecture Slides. Acquired [September
3, 2018]