Imagine Exercise 8: Spatial, radiometric, and spectral enhancement

Objectives:

Instructions:

Open Erdas Imagine 8.3.1, follow the instructions, and answer the following questions.

Spatial enhancement  (see L & K 553-563 for discussion)
1. Click on the Interpreter icon in the Imagine icon panel.  The Image Interpreter menu opens.  Each of the buttons in the Image Interpreter menu will display a submenu of Image Interpreter functions.

2. Select Spatial Enhancement... from the Image Interpreter menu and the Spatial Enhancement menu opens.

3. Select Convolution... from the Spatial Enhancement menu and the Convolution dialog opens.  This interactive Convolution tool lets you perform convolution filtering on images.  It provides a scrolling list of standard filters and lets you create new kernals.  The new kernals can be saved to a library and used again at a later time.

4. In the Convolution dialog, under Input File: (*.img), enter lanier.img.

5. Under Output File (*.img), enter convolve.img in the directory of your choice.  It is not necessary to add the .img extension when typing the file name-IMAGINE automatically appends the correct extension.

Kernal Selection
Next, you must select the kernal to use for the convolution.  A default kernal library containing some of the most common convolution filters is supplied with IMAGINE.  This library will be opened in the Kernal Selection part of this dialog.

6. From the scrolling list under Kernal: click on 3x3 Edge Detect.

7. Click on the Edit... button in the Kernal Selection box.  The 3x3 Edge Detect dialog opens.  For this exercise, you will be using the Kernal Editor to simply view the kernal used for the 3x3 Edge Detect filter.  However, if desired, you could make changes to the kernal at this time by editing the CellArray.  Answer Question 1.

8. Select File | Close from the 3x3 Edge Detect dialog.

9. Click OK in the Convolution dialog.  A job status dialog displays, indicating the progress of the function.

10. Depending on your eml Preferences (under Session | Preferences... | eml | Keep Job Status Box), when the Job Status bar shows 100, indicating that the job is 100% done, you must either click OK to close the dialog or the dialog closes automatically.

11. Select File | Open | Raster Layer... from the Viewer menu bar.  The Select Layer to Add dialog opens.

12. In the Select Layer to Add dialog under Filename: (*.img), click on lanier.img.

13. Click OK to display the file in the Viewer.

14. Open a second Viewer.

15. Select File | Open | Raster Layer... from the menu bar of the Viewer you just opened (Viewer #2).  The Select Layer to Add dialog opens.

16. In the Select Layer To Add dialog under Filename: (*.img), enter the name of the directory in which you saved convolve.img in step 5 and press the Return key on your keyboard.

17. In the list of files, click on convolve.img and then click OK.

18. In the IMAGINE icon panel, select Session | Tile Viewers to compare the two files side by side.

19. When you are finished comparing the two files, select File | Clear from the menu bar of each Viewer.  Answer Questions 2-3.

Radiometric enhancement (read online help)
1. In the Image Interpreter menu, select Radiometric Enhancement...  In this section, you will use both the Inverse and Reverse options of the Image Inversion function to enhance images.

Inverse
2. In the Radimetric Enhance menu, select Brightness Inversion...

3. In the Brightness Inversion dialog under Input File: (*.img), enter loplakebedsig357.img.

4. Under Output File (*.img), enter inverse.img in the directory of your choice.

5. Under Output Options:, turn on the Stretch to Unsigned 8 bit check box by clicking on it.

6. Under Output Options:, click on Inverse.

7. Click OK in the Brightness Inversion dialog to start the process.  A Job Status dialog displays, indicating the progress of the function.

Reverse
8. Select Brightness Inversion... from the Radiometric Enhance menu.

9. In the Brightness Inversion dialog, enter loplakebedsig357.img as the input file.

10. Enter reverse.img in the directory of your choice as the output file.

11. Turn on the Stretch to Unsigned 8 bit check box under Output Options:

12. Click OK in the Brightness Inversion dialog to start the process.  A Job Status dialog displays, indicating the progress of the function.

13.  Open a Viewer and display inverse.img.

14. Right-hold within the Viewer and select Fit Window to Image from the Quick View menu.  The Viewer changes size to bound the image data.

15. Select View | Split... | Split Vertical from the Viewer menu bar.  A second Viewer opens.

16. In Viewer #2, click on the Open icon (this is the same as selecting File | Open | Raster Layer... from the Viewer menu bar).

17. Open reverse.img.

18. In Viewer #2, select View | Split... | Split Vertical to open a third Viewer.

19. With your cursor in Viewer #3, press Ctrl-r.

20. Open loplakebedsig357.img.

21. Resize the Viewers so you can compare the images.  Answer Questions 4-6.

Background Information About Spectral Enhancement: Principal Components Analysis of Digital Imagery
Extensive interband correlation is a problem frequently encountered in the analysis of multispectral image data.  That is, images generated by digital data from various wavelength bands often appear similar and convey essentially the same information.  Principal component transformation is a technique designed to remove or reduce such redundancy in multispectral data.  These transformations may be applied either as an enhancement operation prior to visual interpretation of the data or as a preprocessing procedure prior to automated classification of the data.  The purpose of these procedures is to compress all of the information contained in an original n-channel data set into fewer than n "new channels" or components.    This section has a bit of background on principal components analysis (PCA).  Also read L&K pages 572-577.

PCA is based on the variance and covariance of the data set.
Variance - measure of the scatter or spread within one variable of the data set
Covariance - measure of the scatter between two variables of a data set

The principal components are UNCORRELATED (i.e. orthogonal) even though the original
variables were CORRELATED

Eigenvalues - total variance contained in each transformed axis; the length of an
eigenvector
Eigenvectors - the individual transformed axes; they define the principal components
directions

Principal Components Transformation is constrained such that:
if  l1 = variance of PC1
then l1 > l2 > ... ln

The total variance of the original data set = the total variance of the transformed data set.


 


For correlated data with high covariance:
PC1 > X
PC1 > Y


For uncorrelated data  with no covariance:
PC1 = X = Y

The PCs of UNCORRELATED data are of NO VALUE because no axis may be drawn through a circle that will pass through more data than is projected onto either of the original axes. The elements of the TRANSFORMATION MATRIX are called LOADINGS which are indications of the contributions made to each PC by each original variable. PCA provides a systematic means of reducing the dimensionality of multichannel image data. ITS AIM IS TO PRODUCE DECORRELATED IMAGES. Data in the original channels (bands) are distributed, by means of a linear transformation, between the same number of new channels (the principalcomponents) in such a manner that the newly defined images (PCs) are UNCORRELATED. If there is significant correlation between the original variables, then the first few PCs will contain most of the whole-data-set image information.  The contrast of each PC image is proportional to its eigenvalue (i.e. the total variance contained on that PC); since PC1 contains more variance than PC2 which contains more variance than PC3 ..., the higher numbered PCs in particular require significant contrast stretching in order to make them interpretable.

Thematic Mapper Tasseled Cap Coefficients
Crist and Cicone cound that the six bands of reflected TM data effectively occupy three dimensions, defining planes of soils, vegetation, and a transition zone between them.  The third feature called wetness relates to canopy and soil moisture.


 Table: Thematic Mapper Tasseled Cap Coefficients

TM Band
 1
7
Brightness
 .3037
 .2793
 .4743
 .5585
.5082 
.1863 
 Greenness
-.2848 
 -.2435 
-.5436 
-.7243 
-.0840 
-.1800 
 Third (wetness)
.1509 
.1973 
.3279 
.3406 
-.7112 
-.4572 
 Fourth
-.8242 
.0849 
.4392 
-.0580 
.2012 
-.2768 
 Fifth
-.3280 
.0549 
.1075 
.1855 
-.4357 
.8085 
Sixth 
 .1084 
-.9022 
.4120 
.0573 
-.0251 
 .0238 
The following graphics show how these components (Brightness, Greenness, Wetness, Fourth) can be used in conjunction with one another to interpret a landscape.

TM Tasseled Cap Plane of Vegetation - North Carolina Ag. Farm Data (compare this with next graphic below)

Location of Other Cover Classes in Plane of Vegetation

TM Tasseled Cap Transition (compare this with next graphic below)

Location of Other Cover Zone - North Carolina Ag. Farm Data Classes in Transition Zone View.

 
TM Tasseled Cap Plane of Soils - North Carolina (compare this with next graphic below)

 
Approximate Direction of Moisture Variation in the Plane of Soils (arrow points in direction of less miosture).

(A) Scatterplot of greenness - brightness dimensions of TM Urban Image (compare this with next graphic below)

(B) General postion of surface features in greenness - brightness dimensions.

(A) Scatterplot of greenness - wetness dimensions of of TM Urban Image (B) General position of surface features in greenness - wetness dimensions.

 
(A) Scatterplot of greeness - 4th dimension of TM Urban Image  (B) General position of surface features in greenness - redness (4th) dimensions.

 
Spectral enhancement exercise instructions
1. In the Image Interpreter menu, click on Spectral Enhancement...

2. In the Spectral Enhancement menu, select Tasseled Cap...

3. Under Input File: (*.img), enter lanier.img.  Lanier.img is a Landsat TM image of Lake Lanier, Georgia, obtained by the Landsat 5 sensor.

4. Enter tasseled.img in the directory of your choice as the output file name.

5. Under Output Options:, turn on the Stretch to Unsigned 8 bit check box.

6. Click on Set Coefficients...  The coefficients that display are the standard default entries for Landsat 5 TM Tasseled Cap transformation.  For this exercise you will be using the default entries.

7. Click OK in the Tasseled Cap Coefficients dialog.

8. Click OK in the Tasseled Cap dialog to start the function.  A Job Status dialog will appear.

9. Open lanier.img and tasseled.img in separate viewers.  For tasseled.img under Layers to Colors:, use layer 1 as Red, layer 2 as Green, and layer 3 as Blue.  The image shows a degree of brightness, greenness, and wetness, as calculated by the Tasseled Cap coefficients used:

Answer Questions 7-9.

Q1. Draw a table with the 3x3 edge detect filter values.

Q2. Describe the difference between the two images.

Q3. Using the discussion in L&K as a guide, explain how this filter can be used.

Q4. Compare inverse with the original image.

Q5. Compare reverse with the original image.

Q6. Using the discussion in the online help and L&K as a guide, explain how this can be used.

Q7. Compare lanier.img with tasseled.img.

Q8.  Explain several of the feature classes in tasseled.img.

Q9. Using the discussion in the online help and L&K as a guide, explain how this can be used.