Introduction

In this lab we will learn basic information extraction and image enhancement. We will start with some background discussion on these digital image processing techniques and then walk through some examples within Imagine. 

Background

Information extraction

Information extraction is simply gathering information from the image data. Some simple examples are finding the distance between two points on the image, the area of a particular stand, or the observed near infrared reflectance value for a certain pixel (as you did with the inquire cursor tool in lab 1).

Image Enhancment

Image enhancement is manipulating the image data to create products that are either more visually informative or more appropriate for further image processing. One type of image enhancement with which you are already familiar from the first lab is image reduction and magnification. This is a simple example but makes the point of how you can gather different information from the same image by simply zooming in and out of an image. In this lab we will introduce more advanced image enhancements. The basic idea it to use either global statistics (those from the entire image) or neighborhood statistics (those from a smaller number of pixels) to modify or adjust each pixel's value. Note that after image enhancments are performed, you no longer have the same digital numbers (or 'z' values) for the bands. We will explain this in the lecture and lab introduction.

See Jensen (Chapter 7) for more information on image enhancement.

Some Examples using Imaging Software

Information extraction

Within a viewer open the tm_data/raleigh94_alb.img image. Zoom in on the airport area. Then from the viewer menu bar select Utility/Measure to open the measurement utility tool, which allows you to measure distance and bearing of lines, area and perimeter of polygons or boxes, and other information. This tool can be used to interactively gather specific information from the image. We will now get the length and bearing of the major runway at Raleigh/Durham Airport. From the Measurement Tool Menu, click on the length and angle tool button  and move the cursor to the image.

The left mouse button will start the line and double left-clicking will end the line. Start at one end of the runway by clicking the left button. Move the cursor to the end of the runway and double-click the left button. You should now see some information in the Measurement Tool window similar to this:

We have used the image to quickly determine the approximate length and direction of the runway. This tool can also be used to find the area of a lake, the distance between the airport and NC State, and other information. Of course, this information is not survey quality but does allow the user to quickly get information from a geocoded image.

Image Enhancement

Imagine divides image enhancement into three categories:

1. Spatial Enhancement,

2. Radiometric Enhancement, and

3. Spectral Enhancement.

Spatial enhancement techniques modify the value of a given pixel based on the values of surrounding pixels. Spatial enhancments are also referred to as local operators. Radiometric enhancement techniques modify pixel values based on the range and distribution of the histogram values for the entire image. Recall viewing the histograms in the first lab. Radiometric enhancements will result in modifications in the histogram with the intention of making the image more visibly informative or exploiting the maximum possible radiometric resolution. Spectral Enhancement techniques are mainly transforms that are used to aid in interpreting image data.

These three types of enhancements are available by clicking on the Interpreter box from Imagine's main menu bar to get the Image Interpreter menu.

We will go through the steps of conducting one each of the spatial, radiometric, and spectral enhancements.

An example of Spatial Enhancement

By clicking on the Spatial Enhancement button within the Interpreter box you will give you the following dialog box:

Next click on the Texture button. The resulting dialog box will ask for an input image and the name of an output image.

From the lower right side of the dialog box, you can see that this tool will output either the variance or the skewness of a 3x3, 5x5, or 7x7 moving window. Variance is the average squared difference of the radiance values within the given window. It is a measure of the homogeneity of the data within the window. The higher the variance, the less similar or less homogeneous the data. The Skewness is the average cubed difference of the radiance values. It is a measure of the data's symmetry. The higher the skewness, the less symmetric the data. Either can be used to produce layers that can be added to the original layer to help classify certain areas. For example open water areas are relatively homogeneous, so pixels covering water areas will have low variance values. Conversely, a mixed forest area or residential area will be more heterogeneous, so pixels from these areas will have higher variance values. Using the raleigh94_alb image and the defaults from the Texture dialog box (3x3 variance), create an image called raleigh94_a_tex3v.img.

Open the tm_data/raleigh94_a_tex3v.img as a Gray Scale image, specifying band four as the Display Layer. Add the tm_data/raleigh94_alb.img to the same viewer. By using the Utility/Flicker tool try to determine which land cover types have high band four variance and which have low band four variance.

An example of Radiometric Enhancement

By clicking on the Radiometric Enhancement box within the Interpreter box you will get the following dialog box:

Next click on the Histogram Equalization button. Histogram equalization will redistribute pixel values so that there are approximately the same number of pixels for each data value available. That is, each digital number will have approximately the same number of pixels. Create the image raleigh94_a_histeq.img by using the default values and the raleigh94_alb.img image for input:

Now, open the output image tm_data/raleigh94_a_histeq.img in a viewer. Then, in another viewer open the original tm_data/raleigh94_alb.img image. You should see more visual clarity between features in the histogram equalized image. That is because histogram equalization allows the data values to occupy all possible digital numbers available for a given radiometric resolution.

To see this more clearly, open the Utility/Layer Info tool from each of the two viewers. Once these dialog boxes open, click on the histogram button. By comparing the histograms from the two images, you can see how the histogram equalization enhancement does not exactly make the histogram values equal across all digital numbers but it does spead out the data to take advantage of the entire 0 - 255 range.

Histogram equalization is more for visual display than for subsequent image analysis. Histogram equalization will confuse the physical meaning which is associated with the original digital numbers. By spreading out the data to take advantage of the entire range of digital numbers you loose the original meaning of the digital numbers and their relationship with each other. For example, the extreme outliers from the original data are often grouped together with the upper 1% to 5% of the data. Because it changes the relationship between the digital numbers, histogram equalization is used only for visual purposes.

An example of Spectral Enhancement

By clicking on the Spectral Enhancement button within the Interpreter box you will get the following dialog box:

Next click on the Indices button. Use the raleigh94_alb image as input to create the Normalized Difference Vegetation Index (NDVI) image tm_data/raleigh94_a_ndvi.img:

As noted in the dialog window this index is a function of the original band 3 and band 4, namely:

   band 4 - band 3
   _______________
=
   band 4 + band 3

Note that other sensor types will use different band numbers for the NDVI. The true definition is (IR-R)/(IR+R). By the nature of the index it is limited to values between -1 and +1. High values of NDVI are associated with healthy vegetation and low values with either green non-vegetated areas or unhealthy green vegetation.

The result from running this program in Imagine is a one band image of NDVI values. Open the tm_data/raleigh94_a_ndvi.img image. Add the tm_data/raleigh94_alb.img to the same viewer. By using the Utility/Flicker tool determine which land cover types have high NDVI values and which have low NDVI.

Review Questions:

1. Open a Viewer with both the tm_data/raleigh94_alb image and the doqq/ral_se image. Using the perimeter and area tool , from within the Measurement Tool window, find the area and perimeter of the northernmost of the two athletic fields just north of Jordan Hall on the TM image (use the doqq to help you find this). This area shows up as light red with the default false color composite. Note that for delineating a polygon the left mouse button will start and place nodes along the polygon. Double-clicking the left button will close the polygon. Compare your perimeter and area estimates with others in the class. What would you estimate as the margin of error?

2. Now measure the perimeter and area of the same area on the doqq. Again, compare your estimates with others in the class. What is the margin of error for these estimates? How does the resolution of the image data affect the theoretical accuracy of the information extracted from the image?

3. Try something other than the "Texture" tool from the Spatial Enhancement window. Describe which tool you used, what is does, and how the results of the enhancement could be used.

4. Try something other than the "Histogram Equalization" tool from the Radiometric Enhancement window. Describe which tool you used, what is does, and how the results of the enhancement could be used.

5. Try something other than the "Indices" tool from the Spectral Enhancement window. Describe which tool you used, what is does, and how the results of the enhancement could be used.

6. Why is it important to judiciously use pre-classification image enhancement techniques? When is it necessary to "enhance" an image before performing additional analyses?

Hint for 3,4, and 5: Many of these techniques are described in the Jensen text.

Go back to the FOR 753 Home Page