In this exercise we will analyze an image set to determine what kinds of changes have occurred between
11-84, 9-92, and 10-00. This is an area covering most of Chicagoland. In
this zip folder you’ll find 6 different images from each of these dates named with
the prefixes 11-84, 9-92, and 10-00 respectively, covering TM channels 1-5 and
7. Copy and decompress this folder as we have in the past using 7Zip. Remember, we are at the point of the term where exploration and discovery is important. You
should be sufficiently capable to solve minor problems on your own. Where you
don't understand how a particular module works, look first to the Idrisi help file
and the online manuals (in the Blackboard page), think
about it, and only then ask for help. There are two main approaches to
automated change
analysis:
(1) One might be a simple combination of the three dates of imagery
using color composite techniques
(sometimes preceded by an initial stretch or a radiometric
correction).
(2) Another approach might be to combine registered multi-channel images using
mathematical operators (most normally subtraction or division) to produce new images.
These new images can then be combined with other channels to produce false color
composites that emphasize change, or reclassified using the RECLASS module to set new
threshold values for calculated values in the new images. These kinds of images can have
many uses, including their use as masks for overlay with other images. Remember that you
have a toolkit available to you (e.g. HISTO, EXTRACT, REGRESS, SCATTER, STRETCH, PROFILE,
etc.) to help you analyze these images. Some hints for performing your change analysis:
- In this kind of analysis we generally execute mathematical operations or create color
composites between images in the same channels and different dates (e.g. Channel 1 1984, 1992,
and 2000).
- To perform this type of analysis you must have images that are exactly the same size
(same number of columns and rows) and that are precisely georeferenced. These three
image sets have been created using WINDOW and manipulated using the METADATA module
to produce this result.
- When analyzing urban scenes we most often use the visible channels.
Infrared channels
(i.e. 4, 5, 7) are more useful for detecting environmental/ seasonal change.
- We often want to execute radiometric correction (mainly for sun elevation) to our images
before performing an overlay or a color composite operation, in particular where we are
using images from different dates or times of the year, however, when stretching images
this may not be necessary or recommended. This can be accomplished using the SCALAR
module. This kind of correction is described in the text, pages 499-507. The dates of these
images are November 23, 1984 (sun elevation 24.7 degrees), September 10, 1992 (sun
elevation 45.0 degrees), and October 10, 2000 (sun elevation 38.4 degrees). To
correct images for sun elevation we divide the pixel value (digital number or DN) by the
sine of the sun elevation at the time the image was acquired. Use the SCALAR module
to apply this correction, producing a new corrected image.
- Often when performing SCALAR transformations on images we change integer data into real
data. This can result in failures of particular modules because they are designed to
work on byte/binary data, not real numbers. In these cases you may be required to
use the CONVERT module to convert the file into the data type required by the requesting
module.
- Temporal Image Differencing (text page 597) is a common technique for
change analysis. This can be accomplished using the OVERLAY module to produce a new
image.
- Temporal Image Ratioing (text page 597) is another common technique for
change analysis. This can also be accomplished using the OVERLAY module to produce
a new image.
- After you have used either of these techniques to produce a "change" image you
can then use the STRETCH module to stretch the change image for higher
contrast between values. There are also several useful functions in
TRANSFORM such as Reciprocal, Square, Square Root, and Absolute.
- Another useful and helpful technique is exponentiating a change image. For
example, when you use image ratio techniques, the new image often has new DN's with real
(not integer) values in a range from 0-10. Most unchanged pixels will be right
around 1. If you exponentiate these to the 2nd or 3rd power the contrast between
changed and unchanged pixels can be greatly enhanced.
In this activity you may want to stick with the simple technique of creating color
composites from
either raw, corrected (for sun elevation), or stretched image channels in order to
illustrate change that has taken place between the image dates in 1984, 1992, and
2000. This might make the most sense given the three distinct dates that we have
available. The point of illustrating other approaches to automated change analysis
is to show ways that we might approach this problem if we didn't have such a nice set of
complete images or if we were working with something other than three dates.
When you finish doing your analysis print a D-size version of your change image with a
clear legend (this may be a text document) that explains the color scheme of the image and its meaning regarding
change. Answer the questions below in a separate MS Word document.
1. Describe the process you used to isolate changes that have taken place in this
scene. Be as specific as possible, illustrate using a flow chart, and use
appropriate technical and scientific language.
2. What kinds of change has your techniques proven best in isolating? What
kinds of change are not well shown by your technique and how might you change your
technique to perhaps show this kind of change more effectively?