Lab 7

Radiometric and Spatial Enhancements

Geog 418/518, Fundamentals of Remote Sensing

Name 1 Name 2

Lab Goals

Often raw images do not highlight the particular features or processes that you wish to understand. When this is the case, there are a variety of techniques you can use to show the features more clearly. Techniques used to make images more interpretable for a given application are known as image enhancements. You have already explored some enhancement techniques in an earlier lab where you changed the color of certain features, extracted an AOI (area of interest) that highlighted lakes, forests and urban areas, and worked with 3-D drapes to better visualize the landscape. The goal of today’s lab is to introduce you to two major categories of image enhancement techniques:

· Radiometric enhancement, where you alter the image based on the values of individual pixels in each individual layer

· Spatial enhancements, where you alter the value of a pixel based on its value relative to that of surrounding pixels.

The enhancement techniques you will explore today are techniques commonly used by remote sensing professionals, but are just the tip of the iceberg in terms of the many different approaches that exist (see Table 5-1 from the ERDAS Imagine Field Guide at the end of this lab).

Lab Procedures:

As with all labs, work in pairs on this lab and turn in only one lab per team. Remember to change partners.

Please provide your answers in complete sentences, except where the question specifies otherwise. Please be legible.

Lab Activities and Questions:

Part 1, Radiometric Enhancements.

Radiometric enhancements are techniques that improve contrast between certain features by altering the screen colors assigned to specific ranges of pixel values. For example, if the values of two adjacent pixels are 53 and 57 on a gray scale, and the gray scale varies linearly from 0 (black) to 255 (white), then it will impossible to visually distinguish the pixels on the screen. Both pixels will be an almost identical dark gray.

But if we change the color scale and state that all values from 0 to 50 are black, and all values from 60 to 255 are white, then the values from 51 to 59 can be “stretched” out to display a wider range of grays. In this case, the pixel with a value of 53 will be a medium-dark gray, and the pixel with a value of 57 will be a lighter gray. But this enhanced contrast in the range of 51 to 59 comes at a cost; now we cannot distinguish among the pixels with values of 0 to 50 (which are all black) or the pixels with values of 60 to 255 (which are all white) This is always the case with radiometric enhancements: the gain of contrast in one range leads to a decrease in another.

The following image introduces you to the standard “contrast stretches” available in most any remote sensing software package. Pages 144-154 of the on-line ERDAS Field Guide (1999) provide good explanation of the mathematical concepts behind of each of these stretches.

Go into the class folder for this lab and open up our old friend: lanier.img. But when you open it, check the No Stretch box on the Raster options.

This is a dark image when you view it without an image stretch. In order to understand how to best stretch an image to highlight a specific feature, it is wise to: a) look at the histograms for the raw data to determine the range of data values; and b) determine which data values are associated with which features.

First, reopen the image, but using the ERDAS default settings, which apply a standard deviation stretch that allows you to see the features more clearly

1. Using the Layer/Info tool, look at the general information and histograms for each band. Fill in the table below, collecting spectral data for several water points and several forest points. (Remember that you can use the Spectral tool on the Viewer window to collect spectra for water and forest. Use the View/Tabular option on the Spectral Profile window to give you the values for the bands at each point. You can determine if the histogram for the image values is unimodal or bimodal by going into Utilities/Layer Info from the Viewer window and looking at the Histogram tab.)

=

Band	Minimum value	Median value	Maximum value	Water pixel values	Forest pixel values	Unimodal or Bimodal?
1
2
3
4
5
6
7

2. What two categories of features are driving the bimodal distributions in certain bands? You can use the cursor inquiry tool if you need to determine which features correspond to which pixel values in a given band.

3. Why is the band 6 distribution unimodal, while the surrounding bands (5 an 7) are bimodal?

Imagine provides some television control-like approaches to image enhancement. For example, in the Viewer menu, select Raster/Contrast/Brightness/contrast. Play around with the brightness and contrast to see what happens to your image. Click Reset when you want to return to the original view.

Let’s take a somewhat more sophisticated approach, where you manipulate the brightness assigned to specific data ranges for a specific band. Open up lanier.img again, but only open band 4 using a gray scale. Check off the No Stretch box when you open it.

4. Why is there so little contrast within the non-water and within the water portions of the image? Explain in terms of the raw pixel values (see your table above) and the values these pixels receive on the screen.

Select Raster/Contrast/General Contrast from the View menu. The Contrast Adjust window will open. Play with switching to different methods of enhancement using the Method window. Apply the different methods to see how the image changes. You have to click the Apply button to have the image change.

After looking at some different approaches, apply the Standard Deviation approach.

Click on the Breakpts button. You will get a window that looks something like this (adapted from ENVI Tour Guide, 1999, p. 31):

The graph shows you how the original pixel values are adjusted for display purposes using the Standard Deviation approach. For example, a pixel with a value of 54 will now be given a screen value of 148. If the original image was displayed with values of 0 being black, values of 255 being white, then the value of 54 would be a medium-dark gray, which would be very similar in tone to all the other pixels with values of 40 to 60, which make up most of the pixels in the graph above. The new screen value of 148 will be brighter. More significantly, pixels that previously had values close to 57 are now “stretched” out over a wider range, as is shown by the output histogram shown in red. l

In the Contrast Adjust window, change the Standard Deviations from 2.0 (the default) to 1.0 and click Apply.

5. How does the breakpoint curve in the Breakpoint Editor window change? Does it become steeper or shallower?

6. Describe how the output histogram (the yellow one on your window) changes when you go from a standard deviation of 2.0 to 1.0. Specifically, does the 1.0 std dev output histogram have more of fewer values in 240 to 260 range? Why is this?

Click on the Apply button in the Breakpoint Editor window and see how the contrast changes in the Viewer window for lanier.img when you are using a 1.0 std dev stretch.

Now play a bit with the breakpoints. You can click and drag the existing breakpoints, or click on the Insert Breakpt icon in the Breakpoint Editor window and inserts some more breakpoints. Try dragging the breakpoints, or inserting some new breakpoints and examining how the image changes. Remember, you have to click on Apply to see how this affects your image.

You can insert breakpoints to highlight a certain feature if you know something about the original feature.

7. Based on your table above, and assuming that the range is a it wider than the pixels you found, what range of values in band 4 would you want to stretch in order to highlight variations in the water?

Click on the arrow icon in the Breakpts Editor window and drag the breakpoints so that the minimum and maximum values fit the range your just stated for water values. Click apply to see how it changes the image.

Zoom in on the lake and surrounding shore.

8. What color is everything outside the lake? Why is this?

If you have been successful, you will now have a variety of shades of gray inside the lake.

9. Where are the darkest pixels in the lake area and why do they occur in these locations?

10. Where are the medium and lighter gray pixels located in the lake. Why do they occur in these relative locations

Now repeat this process, but with range of values you outlined for forest features for band 4.

11. Now what shade is the lake and why is this the case?

12. Examine the forest near the shoreline. What is the range of tones you see now in this region?

The above exercise hopefully gives you a sense of what you can do to visually highlight key features of interest. Many folks are inclined to indiscriminately apply stretches – and sometimes that can work. But without too much extra effort, you can gather some spectral information about the key features of interest as you did above, and then use that information to create custom stretches that highlight exactly what you want to highlight.

Feel free to play around with image stretches. In particular, you might want to bring up the image as a three band composite and alter the stretches while viewing the image in color.

Part II, Spatial Filters

For this part of the exercise, let’s work with a real world example for which I have not yet found a satisfying answer. A student in class last year brought me this problem as part of a project she was doing with the city of Eugene. The goal was to use a scanned black and white photo of the Spenser Butte area to map trails. If we are fortunate, one of you may come up with the ultimate trail map that Bethany and the city of Eugene would like to have. I certainly haven’t come up with any brilliant maps yet.

Note: Each of the images you generate in the following exercise will be about 5 meg in size. You need to clear out any old images you have stored in your folders to avoid exceeding your space limitations.

Open up Spenser.tif (note that it is a .tif file – you will have to tell Imagine to look for tiffs when you open the files). Zoom in on Spenser Butte and the clear cut to its south. At a quick glance, only the most obvious trails in clear areas can be seen on the photo and certainly none of the trails in the forest come out clearly.

I spent some time playing with radiometric enhancements (i.e., contrast stretches), just as you just did in the previous part of this exercise. The contrast stretches did enhance some of the trails. In particular, I had some success with a linear stretch where I went in and inserted some more break points to stretch out the bright reflectance values (on the assumption that a clear trail will be brighter than the surrounding vegetative surface).

For this lab, however, let’s take a look at some spatial enhancements, also called spatial filters. Spatial filters are described in the text and also on pages 154-160 of the ERDAS Imagine Field Guide, 1999. Spatial enhancements look at the value of a pixel and its surrounding pixel, then applies equations that either increase or decrease those differences. Enhancements that increase those differences are known as high pass filters and can be particularly useful for detecting edges (like trails). Enhancements that decrease differences are known as low pass filters and are useful for removing noise, merging.

Click on the Image Interpreter icon

(one could also reach the following set of commands through the Raster/Filter options on the Viewer window). Select Spatial Enhancement, then click on Convolution.

Enter spenser.tif as the input file, and enter spenser-3edge as the output file. Chose the 3x3 Edge detect kernel and click OK.

Repeat this process using the 5x5 edge detect kernel and the 7x7 edge detect kernel to create two other files: spenser_5edge and spenser 7edge.

Finally, create a file named spenser_nondirection by clicking on the Non-directional Edge option in the Spatial Enhancement window. Select the Prewitt option for your enhancement.

Open your four edge detection images in separate views. .

13. Which of the four approaches, if any, do the best job of picking up trails? Zoom in on the area around Spenser Butte to check this out.

14. Visit the NE corner of the images. Which image does the best job of mapping the housing and roads in that area?

15. Given your answer to 14, what might be an application of edge enhancement detections for mapping purposes?

16. How would you describe the differences between the 3x3, the 5x5 an the 7x7 edge enhancement figures in terms of texture, the size of objects you can discern on the images, and the range of grays you see on the image. (Zoom in on the image and look at the histograms for the images if you are having trouble coming up with an answer here.)

Table 5-1 from the ERDAS Imagine Field Guide, 1999, p. 138-139.