Lab 5
Highlighting
features of interest on an image
through
use of band combinations, colors, AOIs, and DEMs
Geog 418/518, Fundamentals
of Remote Sensing
Name 1 Name 2
The goals of this lab are to:
a) to introduce you to basic approaches for highlighting features of interest on an image;
b) to better understand the data behind the image; and
c) to develop some software skills that you will use throughout the remainder of the quarter and are likely to use in any remote sensing project.
The techniques you will explore in this lab all fall under the category of activities that remote sensing specialists call “image enhancement.” With image enhancement, you alter the appearance of the image or subset the image in a way that highlights whatever feature or features you are most interested in. In some cases, the feature of interest may be the whole image and your primary goal will be to highlight the contrast so that key features show up more clearly. In other cases, you may want to highlight a specific place or a type of feature (e.g., coniferous vegetation). In all cases, image enhancement is done in order to better visualize the features of interest.
The image enhancements we will explore today change how digital data is displayed on the screen, but do not alter the underlying data. Other image enhancements that we will examine later in the quarter (e.g., principal component images, smoothing and sharpening algorithms, etc.) do change the data in order to highlight specific features.
As with the previous lab, some of the commands will be provided to you in the text below, but other commands or information are not provided. I expect you to try searching the Help menu and playing with various alternatives before asking for help from me or the GTF.
General background information on remote sensing and data storage topics can be found in the ERDAS Imagine Field Guide. Specific help on the software and which buttons to push to access certain pieces of information or commands can be found in the ERDAS Imagine Tour Guides. Both guides can be accessed via the Help button on the ERDAS menu.
Lab Procedures:
As with the last lab, work in pairs on this lab and turn in only one lab per team. This week, however, change partners. Switching partners gives you a chance to get to know more of your classmates and introduces you to a wider variety of learning styles typical of future colleagues you will encounter in the remote sensing world.
Please provide your answers in complete sentences, except where the question specifies otherwise. Please be legible.
Lab
Activities and Questions:
Go into the lab folder for this exercise and open lnsoils.img
Raster images can depict either continuously varying data, where each pixel can have a separate value (e.g., a Landsat image), or discrete classes (e.g., a soils map), where many adjacent pixels have the same value. When areas of the landscape are lumped together into classes, we call this a chloropleth map.
At one level, displaying a map of this sort as a raster image doesn’t make much sense; one would usually portray classified data in a vector-based program for cartographic purposes. This is because: (1) vectors are better at showing boundaries precisely; and (2) because it is inefficient to use rasters to store data when the same value covers large areas. For example, if the spatial resolution of an image is 10 m and there is a cluster of soil type 2 that covers 10 hectares, then one would have 1000 pixels adjacent to each other, all with a value of 2. It is generally more efficient to use a vector to draw a boundary around the area and then assign a value of 2 to the entire area. In this case, the amount of space required to store the attribute data (i.e., the value 2 that describes the soil type) is 1000 times less for the vector system than the raster system.
In some cases, however, one may wish to keep the discrete class data in raster form. You might do this so that you could easily overlay the classes with a raster image, so that you could analyze the relation of classes (e.g. correlate the pixel values between different images), or simply for convenience because you don’t want to covert images back and forth between vector and raster formats.
When classified data is stored in raster form, you may wish to adjust the color schemes used to show the different classes. Just as with map making programs, you can do this by adjusting the palette for the entire image. The following steps and questions address the issue of altering the color scheme to improve appearance. The sequence is loosely adapted from the ERDAS Imagine Tour Guide.
In the Viewer menu bar, select Raster /Attributes. Column 1 of the Raster Attribute window shows you the pixel value assigned to each class. Column 2 shows you the class name, and column 3 shows you the color. Many more attributes are displayed in subsequent columns. Take a moment to explore the columns and their content. Also, click on the Edit heading on the Raster/Attribute window and look at the various options that are available to you. They are many ways to export or alter the classes and image data. For now, however, we are only concerned with changing the color.
You can select a class by clicking on the row number or clicking directly on the class name. Select the water class and right click on the color for that class.
A number of alternative colors are displayed. You may choose from any one of these, but let’s get more exotic in our color choices. Click on Other. The Color Chooser window has now opened up.
You can alter the colors and their intensity by:
a) dragging the dot on the color wheel.
b) Adjusting the numbers on the color number fields
c) Moving the color slider bars.
d) Clicking the Standard tab, which provides a list of predefined colors that work well for different purposes.
In all cases, you must click Apply to have the color show up in the View window. If you wanted to keep the new color, you would have to save it.
Play around with varying the colors in different ways. Learning to create a palette that is both pleasing to the eye and more effective at conveying information is key to “selling” your remote sensing results to clients and users.
As you look at the color number field, you will notice that
there are numerical boxes for red (R),
green (G), and blue (B). These numbers
represent the proportion of the color that is made up of each primary color,
ranging form 0% (0) to 100% (1.0). By
altering the numbers in the boxes, or moving the slider bar to their right, you
can combine these colors in different proportions to produce any of the colors
we can distinguish with the naked eye.
Test how combinations of these primary colors can be combined to create
other colors.
Note that the vertical
slider bar to the right of the wheel controls the intensity (brightness) of the
color. Play with this to see how it
affects the color.
1. Create the colors listed below, then fill in the table with the numerical values showing what proportion of each primary color is needed to create that color.
|
Color |
Red |
Green |
Blue |
|
White |
|
|
|
|
Black |
|
|
|
|
Yellow |
|
|
|
|
Medium gray |
|
|
|
|
Cyan |
|
|
|
Now click the IHS tab on the Color Wheel window. Play with the slider bars that control the intensity (I), hue (H), and Saturation (S).
2. In your own words, explain what it is that the intensity, hue and saturation refer to.
Finally, take a look at
the color scheme and the number of soil classes shown on the lnsoils.img.
3.
From the perspective of a user trying to
identify areas covered by different types of soil, what is a problem with this
map and how might you fix it?
Open the lanier.img file
assigning bands 3, 2, 1 to the R,G, B screen color. Make sure to the Raster/Options is set so
that it does NOT clear the display (i.e., it should leave both the lnsoils.img
and lanier.img files open at the same time).
Go into the Utility command on
the View windwon menu and experiment with Blend, Flicker, and Swipe. (This is not the name of a rock band,
although Blend, Flicker and Swipe does have a certain ring to it).
4.
How might use of one or more of these tools help
to to visually interpret the image layers and their relationship?
Part 2. Altering
band combinations to highlight key features
Close the lnsoils.img
file. If you do not already have it in
the Viewer window, open lanier.img and assign bands 3, 2, 1 to the R,G, B
screen colors.
5. How well can you distinguish variations within the forest with the 3,2,1 band combination? From the perspective of scattering, reflectance, and/or absorption, why is this the case?
6. Based on your answer to 5 above and your knowledge of reflectance/scattering/ and absorption, what might you do in terms of band selection to enhance your ability to see (to visualize) differences within the forest? Explain your reasoning for your answer.
Zoom in on the lake/forest boundary so that you can see individual pixels
7. How well can you distinguish the boundary between the lake and surrounding forest on this image? In terms of scattering, reflectance, and/or absorption, why is this the case?
8. Based on your answer to 7 above and your knowledge of reflectance/scattering/ and absorption, what might you do in terms of band selection to enhance your ability to delineate (to visualize) the lake/forest boundary? Explain the reasoning for your answer.
9. Would you use band 6 to delineate water from vegetation? Why or why not?
On the View window,
select the Raster/Band combinations
command and reset the bands the way you suggested in 6 and 8 above. Did your suggested changes in the band
selection improve your ability to visualize differences? If not, try some other band combinations to
see which one works best for the purposes outlines above.
A key part of remote
sensing image interpretation is selecting the correct band combinations. This relatively simply step can make a great
difference in your ability to visually interpret an image.
Part 3. Defining Areas of Interest (AOIs)
Often you want to
highlight or cut out a certain part of an image. This area might be defined in spatial terms
(e.g., the areas within a box, the area to the east of the highway, etc.) or in
spectral terms (all adjacent areas that have a reflectance like water). Highlighting or cutting out these areas can
be a key part of visualizing the landscape.
Make sure that lanier.img
is open in the Viewer window. On the
Viewer window menu, select File/New/AOI layer. This tells the system that it is
about to create an area of interest.
Check your View/Arrange layers
command to make sure the AOI layer has been established.
Select AOI/Styles from the Viewer window menu and explore the
commands to adjust the box colors and fill patterns. Be sure the Fill box is checked on. This can be crucial to being able to see the
AOI on the screen. Close this window
when you have the properties you desire.
(My preference is to use yellow outlines (“chasers”), which show up well
against most remote sensing image color combinations.)
As a starting point,
let’s try to highlight the lake as your AOI.
To accomplish this, we will “plant” a pixel “seed” and ask the system to
“grow” the region around the seed by looking for adjacent pixels with similar
reflectance signals. This is a spectrally defined AOI, as contrasted with an
AOI created by simply drawing a polygon around the areas we want highlighted.
Select AOI/Tools from the Viewer window menu. Click on the Seed tool, which looks like a little magnifying
glass. Place the cursor in the middle of
the lake and click.
Hmmm…. Chances are, you
don’t see anything happen. This is because
your “seed” is so specialized it can’t find any similar pixels around it. Zoom in on your seed and see if you can make
it out. Probably one or two pixels have
been highlighted. This is your AOI. But it is not a very useful AOI for capturing
the lake. To capture a larger area,
let’s change the spectral properties so the seed does a better job of cutting
out the lake.
Select AOI/Seed properties from the Viewer window menu. This dialog sets the controls on how a
spectral AOI is created. The Euclidean
spectral distance refers to how similar adjacent pixels are in terms of their
reflectance values. Low values refer to
very similar pixels (i.e., the spectral “distance” between them is small). High values refer to dissimilar pixels (the
“distance” is great).
Play with the controls by
changing the spectral distance. The
default is 1.0. Change the value to 2.0,
then 3.0, and so on.
10. What value for the Euclidean spectral distance is best at outlining the lake while also avoiding having “holes” in the middle of the lake?
Even after choosing a
better spectral distance, the AOI does not cover all of the lake. This is because the default is set to create
an AOI no larger than 1000 pixels. Reset
the number of pixels box on the AOI Seed Properties window to 10,000 pixels and
hit the Redo button.
11. How does the AOI change?
Now zoom out so you can
see the entire image. Explore different
combinations of pixel numbers and Euclidean spectral distance until you have a
combination that outlines the lake fairly effectively. Set the number of pixels to 25,000.
12. What minimum value of Euclidean distance outlines all the lake (including bridges) without including other features? Provide the number.
Delete the AOI you just created (use the scissors symbol on the AOI Tools window) and explore the creation of new AOIs for urban areas and forest. See how these AOIs are affected by altering the Euclidean spectral distance.
13. Do you
believe your
The exercise above had you
develop a spectrally-based AOI.
Sometimes, this is all that is needed to identify areas that are
spectrally similar and deserving of investigation. I have used this simple technique, for
example, to highlight areas of algae along a river that were potential frog
habitat. This is far simpler and often
just as effective a visual approach as some of the classification techniques we
will examine later in the quarter.
If you wanted to, you could also draw your own boundaries using the box, ellipse, point, line and polygon tools in the AOI Tool window. Take a moment to use these tools to try to create an AOI for a portion of the lake.
14. Do you believe the spectrally-based or the hand-drawn AOI was better for delineating the lake? Why is this the case?
15. In what circumstances might you use a spectrally defined AOI? In what circumstances might a hand drawn AOI be the better option?
If you were going to use
these AOIs at a later time, you could save them using the File/Save command. We do not need to do this today, however, so
you can close the lanier.img file without saving the changes.
Part 4. 3-D displays
Finally, one of the
favorite visualization tools used for enhancement purposes is the generation of
3-D displays.
Open the eldodem.img and
eldoatm.img files in your Viewer window.
Select Utility/Image
drape.
An Image Drape window will open up with the image draped on top of the
DEM.
Note that a line
connecting the “Eye” and the “Target” will show up in your Viewer window. Explore moving the Eye and the Target around
to see how the perspective in the Image Drape window changes. You may have to select the View/Update command in the Image Drape window after you
reposition the Target or Eye.
After you have a feel for
the ways you can reposition the Target and Eye, select the icons in the Image
Drape window and explore how you can vary sun angle and perspective.
16. As you look from SW (the left bottom of the image) to the NE, which sun angle works best for visualizing the landscape? Why is this?
3-D perspectives can be
great fun and are excellent for PowerPoint presentations, but are not used a
great deal for classical remote sensing studies. Outside of simple presentations,
the most common and valuable use for these 3-D types of displays is the
determination of View Sheds, which are images showing all the sites that can be
seen from a particular vantage point.
View shed calculation is a powerful tool for land use planning, because
it allows planners to place unsightly activities out of sight of the majority
of the population or local users.
Timber, mining, and ski operations now use this tool extensively.