Digital Image Analysis

Exercise 4 and 5: Image Processing and Classification
Using Easi/Pace (Xpace)
The purpose of this exercise is to get an introduction to digital satellite imagery using the software
EASI/PACE ImageWorks and to learn about basic image processing functions.

Image processing
Introduction
Go to your directory by typing the unix command cd gr* in the terminal window, where * is replaced
by your group number. If you type ls on the command line you will get a list of the files stored on your
directory. The file you will work with in this exercise is called savlamar.pix.

Start the software Xpace by typing Xpace on the command line.

If you have several windows on the display, click on the one you want to use and it will be activated.

ImageWorks
We will begin with opening ImageWorks.

1. On Xpace upper menu bar you will find Image Display Windows under Utilities.
Activate it and in a new window you will see a list of Image Windows. Choose the first Image
Window on the list with a mouse click on that line. Click on Modify and change the number of
image planes to seven (7C), if that is not already the case. To start click on Start ImageWorks.
Two windows will be created, a display window and a control window. Move the windows so that
you see both.
2. On the display menu bar there is a File option. Under this you will find Load Image....
In the next appearing window, choose savlamar.pix. A new window appears were you select the
channels and the part of the image you want to load to the display. Select all the channels with
Default and the entire image with Overview. Press Load&Close. After a while a dark image will
appear in the display window.

Help gives more information about the control panel and ImageWorks.

Enhancements
3. Explain why the image is so dark on the display by looking at the grey level histogram for the
unstretched image under View → Histograms....
To make the image more agreeable to look at, we use the enhancement capabilities in the ImageWorks
control window. There are 5 types to choose from: None, Infreq, Linear, Equal, and Root. Try
Linear, Equal, and Root. Look at the differences of the different techniques.
If you open the Look Up Table, LUT... under Edit, you will be able to see the effect on the data
in the histogram for the different enhancement techniques. Describe them. Which method do
you prefer? Why?
4. Also try to zoom in the image after making a stretch on the whole image, and then use the same
stretch again. To zoom in to a specific area, first place the cursor in the image at the point you
want, and then zoom.
Centre for Image Analysis March 14, 2005 2

Is there a difference between the result of the enhancement compared to when you did
not zoom? If so, describe the differences, and why there is a difference.

Grey level values and histograms

5. There are three different ways to display the images. In the control window there is a button where
you can choose between RGB, BW and PC. RGB = Red, Green, Blue and by assigning one channel to
the red image layer, another to the green, and a third to the blue layer you can look at a true or
false coloured image. In black and white (BW) you look at one channel at a time. Pseudo colour
mode (PC) is used for, e.g., classified images.
Try some RGB combinations of your image. You can find information about the image channels
under View → Image Info.... Which combination is the best for showing a true coloured
image? Why is this the best one?
6. To see what the grey level values are in each pixel, move the cursor over the image. In the upper
part of the control panel you can see the image coordinates for the pixel, and the grey level value
for the channels you are looking at.
What are the grey level values of pixel 200,200 for the true colour image?
An other possibility is to use Numeric Values... under View in the image display upper menu.
Try also this.

The image. . .
7. The image you are looking at is a Landsat TM image showing the coastal zone of southern Jamaica.
The spatial resolution of this image is 28 meters. What is meant by spatial resolution?
To see the image coordinate for each pixel, move the cursor over the image and look at the numbers
in the upper part of the control panel. How many pixels does one image line/column consist
of ? How many kilometers does that correspond to? Do you think the image you are
looking at is a complete scene recorded by the sensor or just a part of it?
8. How many bands does the TM (Thematic mapper) sensor record the reflected light in?
To know more about the spectral characteristics of the bands in this image, look at Image Info...
under View. How many image bands does this image have? How many of them are in
the visible part of the spectrum?
Look at the different bands by displaying them in black and white one by one.

Filtering
9. Open Filter... under Tools. Two windows will appear; a small filter pre-view window and a
control panel.
Start to select Input Image as, e.g., 3 and Output Image as 6 (band number 6 is empty and can
be used for results). Mask should be Entire Image. In the ImageWorks control panel you should
select one band to show in BW (preferably the one you choose as input image). You are now free
to try different filters and filter sizes. As soon as you change, e.g., the filter type, the result will be
seen in the Filter Preview window.
If you press Apply, the result for all of the image will be seen in the output image (6).
Two problems to look at:
(a) In the water, you can see that the images are striped. The striping varies a bit between the
channels. This is a result of a defect in the satellite sensors, and the calibration. Try to reduce
the striping in channel 3 or 4 with, e.g., Average, Gaussian and Median filters of different sizes.
You can read more about each filter if you press Help.
Which filter gave the best result? How and why?
(b) In the top left part of the image you can see roads between the fields. Try to enhance the roads
with some of the edge detection filters.
Which filter gave the best result? Use Help and try to figure out how that filter
works.
Centre for Image Analysis March 14, 2005 3

Classification
The programs that will be used in this part of the exercise, called classification, are included in the Xpace
software or written in EASI/PACE macro language. The procedures consist of many programs that can
be used in a step by step image classification. The macros that you will use are in most cases self instruc-
tive.

The classification will be performed on landsatTM.pix over Uppsala.

The Training Dataset

The training dataset is a 512 pixels training area cut from a Thematic Mapper scene from 1989 07 07. It
covers a region (12.5 × 12.5 km) east of Uppsala län. A topographic map (11I NV) and an image plot will
help you to identify different features in the image.

The dataset used for the exercise called landsatTM.pix contains the following items:
• 7 Channels with Landsat TM band 1,2,3,4,5,6 and 7 data TM 1 and 6 will not be used in the
classification.
• 2 Channels to use for results.
All this information can be found in the PCIDSK File Browser... under Utilities in the Xpace win-
dow. Click on New file... and choose landsatTM.pix, to see the information abou that file.

Open landsatTM.pix in a display window using following procedure:

On Xpace upper menu bar you will under Utilities find Image Display Windows.... Activate it
and in a new window you will see a list of Image Windows. Choose the first Image Window on the list
with a mouse click on that line. Click on Modify and change the number of image planes to nine (9C), if
that that is not already the case and the size to 512 × 512 lines and columns. Then, to start the window
click on Start ImageWorks. Two windows will be created; a display window and a control window. Move
the windows so that you see both.

On the display menu bar there is a File option. Under this you will find Load Image.... In the
next appearing window, choose landsatTM.pix and press Ok. A new window appears where you select
the channels and the part of the image you want to load to the display. Select all the channels with
Default and the entire image with Overview. Press Load & Close.
Centre for Image Analysis March 14, 2005 4

Examine the Database

Image control panel

Cursor pixel and line coordinates

DN values for selected bands

RGB settings

Graphics on/off
Graphics shown on display

Vectors on/off
Vectors shown on display

Enhancements

Image zooming

The image displayed contains 7 channels. The image can be stretched by clicking on one of the
Enhancements button in the image control window. Look at the seven channels separately by choosing
Imagery BW and then on each channel. Different enhancements can be applied either by using the zoom
option in the handler and zoom the area to enhance or by using Edit → LUT... option on the display
menu.

channel 1 is Landsat TM band 1 0.45 − 0.53 µm

channel 2 is Landsat TM band 2 0.52 − 0.60 µm
channel 3 is Landsat TM band 3 0.63 − 0.69 µm
channel 4 is Landsat TM band 4 0.76 − 0.90 µm
channel 5 is Landsat TM band 5 1.55 − 1.75 µm

channel 6 is Landsat TM band 6 10.40 − 12.50 µm

channel 7 is Landsat TM band 7 2.08 − 2.35 µm

Examine these channels and answer the following questions:

1. Which channel shows the best contrast between water and land ?
2. Look at an area with houses and roads and describe the difference in the different wavelength
bands.
Enable the Imagery RGB on the control window.
Centre for Image Analysis March 14, 2005 5

3. Examine different band combinations by selecting which band to apply to each colour.
TM 3,2,1 means TM3 in red, TM2 in green and TM1 in blue, corresponds well to the visual colours,
a true colour composite. Compare this band combination with TM4 in red, TM3 in green and TM2
in blue. Can you give any comments on the advantage and disadvantage of using TM4
(infrared)...?
4. The combination TM 5,4,3 is used quite often, can you see any differences or advantage
compared to the usual IR aerial film combination like TM 4,3,2 ?
5. TM6 is very different from the other bands. How and why?

Classification and image segmentation

The simplest way of performing a classification is to do a segmentation in one dimension (one band/channel).
You could for example say that all pixels between DN 5 to 23 in band TM4 belongs to the class water.
Water is one of the few classes that can be sufficiently separated using only one band. (If you want
you can try it using the program THRESH in the Xpace window. Ask the teacher for explanation of the
program.) Most of the time you will need two or more bands to identify a certain objects, for example
differents kinds of forests or agricultural fields. We will start the classification session by performing a
two dimensional classification and then continue with a multispectral classification where we will use more
than two bands.

Two-dimensional classification
The scattergram
By plotting image grey level values against each other we get a 2D-graph called a scatterplot. Each
scatterplot is a combination of the histogram of two channels showing the correlation between these his-
tograms. The brightness of each scatterplot pixel is determined by the number of pixels in the image
with the corresponding grey level values. Thus, bright areas indicate common combinations and black
areas indicate combinations that virtually never occur. Use the topographic map and image plot Upp-
sala NV in scale 1:50 000 to find training areas and to identify features. “Gröna kartan” will be handed out.

Start the program by clicking on the CLASS2D button in Xpace.

There are 4 buttons under the CLASS2D program. RUN, ABORT, FILE and DBIC.
DBIC is used to select which channel to use for the scattergram.
FILE is used to select the file to use in the classification.
Use Landsat TM3 and TM4 for scatterplot by entering 3,4 in DBIC or by clicking on the rows showing
those channels. When two channels are chosen you can run the program by pressing RUN. The HELP
button will give you more information. Wait. . . . . .

Select the I option in the menu that is shown. In the ImageWorks display menu bar there is a graphics
editing option (Edit → Graphics...). Activate this option and it will give you a menu for drawing and
editing graphic overlays.

Select graphic number 3 (blue) in the graphic editing menu.

Select your training area for water in the image and check this with the map.

Outline a training area in the image by using the graphic editing function Trace & Close. (There
are other options that you are free to try.) Fill with the Flood Fill option. Use Help if you need an
explanation of the different tools.
When you are ready with your training area, move back to the CLASS2D menu in the terminal window
and press return to continue.
The training area will now be plotted in the scatter plot in magenta. The magenta coloured pixels repre-
sent the training area in the scatter plot. You can now run the classification by using the training area
cluster to mark out all pixels that match the grey level value set of TM3 and TM4.
Enter y to run a quick classification. Wait. . . . . .
Now you can edit the training area in graph 1 (red) in the graphic editing menu. Remove pixels far away
Centre for Image Analysis March 14, 2005 6

from cluster center and fill out holes. The zoom function can be used as usual.
Enter return in the CLASS2D menu when you are satisfied with your editing.
Classification will be shown in green. You can compare the image and the classification by entering B and
save it with G.

To enter a new training area make the same procedure again but start with deleting the old one. This can
be done by pressing the button number three in the Graphic Clear part of the graphic editing window.
Examine the other options and answer the following questions:
6. Is it possible to separate different lakes (Ekoln and Funbosjön) and in that case why?
Is it possible to outline those in the scattergram with the S option ?
By visual inspection, what other classes is possible to separate.
Built up areas?
Coniferous dense forest?
Different kinds of crops?
Clear cuts?

Multispectral Supervised Classification

This is just a very summarised description of the steps in the supervised classification. Look at the pages
from the manual for more information. Compare your obtained classification with “Gröna kartan” from
the area covered by the image. The manual and “Gröna kartan” will be handed out.
1. Open landsatTM.pix in a display window.
2. Start by clicking on Supervised... under Classify.
3. A small toolbox is opened up. First press Channel Setup and choose New session.
(a) Classification input: Choose whatever bands you would like to use for the classification.
In your report you should comment your choice.
(b) Training sites: Use the band designed for it by the system: 9
(c) Visual reference: Choose any three band for display. In your report you should comment
your choice.
4. Go to second step, Load new area. In this particular case you might have just the whole image
already displayed, which will give the same full resolution result.
5. Signature editing is the next step. A new menu is opening up. Use any (short) Name and then
in Description give a more detailed explanation. You may leave the number as it is or choose any
other.
Whenever you are satisfied with what you have done you can save it by pressing update. Then you may
choose New signature. The rest is probably self-explanatory.
6. There are a number of tools available to better generate the proper signatures. USE THEM!! Look
in the manual for better instructions on the tools.
You may any time check your classification in the preview. You can make a scatter plot to see the
separability between classes or make a histogram. There is also signature statistics to help you and
signature reparability.
7. Classification to file: In real cases you must be aware that you need one channel for the Train-
ing set and one for the Result. Those channels must be created before you start the classification.
Here those two channels are already created for you. Channel 9 is used for the training set and you
should use channel 8 for the classification result. Press channel 8 in the upper part of the menu.
The lower part is used to present the classification statistics to you.
You should save the statistics and give comments on it to the teacher in your report. How
well are the training pixels classified? What is the overall accuracy?. . . . . .

Print the classified image. First Print under File to a .ps-file from the image window. Then use
UNIX command: lpr -Pclc700 filnamn.ps.