Remote Sensing - Lab 4: Miscellaneous Image Functions

-

Goal and Background

The goal of this lab exercise is to learn about and practice using the various image functions that can be found within ERDAS Imagine, which will be beneficial for future applications of image analysis.

Lab 4 is split into seven parts, with the goals of each part outlined below:

Part 1: Image Subsetting (Creation of Area of Interest AOI) of Study Area
Goal: Subset satellite images of Eau Claire and Chippewa Falls area

Part 2: Image Fusion
Goal: Use the 15m panchromatic image to ‘pan-sharpen’ the 30m reflective image

Part 3: Simple Radiometric Enhancement Techniques
Goal: Perform preliminary radiometric enhancement techniques to enhance image spectral and radiometric quality

Part 4: Linking Image Viewer to Google Earth
Goal: Link ERDAS Image Viewer to Google Earth

Part 5: Resampling
Goal: Resample images based on analytical need

Part 6: Image Mosaicking
Goal: Perform an Image Mosaic on a study area that traverses multiple satellite image scenes

Part 7: Binary Change Detection (Image Differencing)
Goal: Estimate and map the brightness values of pixels that have changed in Eau Claire county and four other neighboring counties between August 1991 and August 2011

Methods

The methods used to reach the goal in each part of Lab 4 are listed below. All resources used were provided by the instructor and can be seen in the Sources section at the bottom of this post.

Part 1: Image Subsetting (Creation of Area of Interest AOI) of Study Area

Goal: Subset satellite images of Eau Claire and Chippewa Falls area

There are two general methods that can be used for subsetting satellite images. (1) Using the Inquire Box to create a rectangular or square box within the satellite image scene or (2) Using a headsup digitization or a shapefile to delineate an Area of Inerest (AOI). 

 (1) The Inquire Box is added to the image by right-clicking the mouse and selecting 'Inquire Box'. The box appears on the image and can be moved around or resized to cover the area of interest. To generate the subset image from the Inquire Box, select the Raster tools and click the tool Subset & Chip and Create Subset Image. The output file will now include only the area of the image that was inside the Inquire Box.

(2) It is also possible to delineate an AOI by using a shapefile. Add the shapefile onto the viewer so that it overlays the original image. Select the shapefile(s) and click the Home tab and then Paste From Selected Object, which will create an AOI around the shapefile, represented by dotted lines. Be sure to save the AOI, which will be used in the Subset & Chip and Create Subset Image tool.

Part 2: Image Fusion

Goal: Use the 15m panchromatic image to ‘pan-sharpen’ the 30m reflective image

In order to optimize the original reflective image for visual interpretation purposes, the higher spatial resolution panchromatic band can be combined with coarse resolution image in a method called 'pan-sharpen'. The panchromatic image has a spatial resolution of 15m, which means the image quality will be much better when compared to the reflective image, which has a spatial resolution of 30m.

How to 'pan-sharpen' an image:

Open two viewers, each containing the images that will be combined for the 'pan-sharpen'. Click on the Raster tab and then select Pan Sharpen > Resolution Merge. The Resolution Merge window will open where the panchromatic image should be selected as the "High Resolution Input File" and the reflective image selected as the "Multispectral Input File". Under Method, select the "Multiplicative" radio button and use the "Nearest Neighbor" Resampling Technique. Click OK to run the resolution merge model. Open the new output image and link views with original reflective image to visualize the changes that were made with the 'pan-sharpen'.

Part 3: Simple Radiometric Enhancement Techniques

Goal: Perform preliminary radiometric enhancement techniques to enhance image spectral and radiometric quality

How to perform a Haze Reduction on an image:

With the image open in a viewer, click the Raster tab and then select Radiometric > Haze Reduction. Select the image to be enhanced as the input file. Accept all the default parameters in the window and click OK to run the model.

Part 4: Linking Image Viewer to Google Earth

Goal: Link ERDAS Image Viewer to Google Earth

The ability to link images to Google Earth is a new development in Erdas Imagine 2011. This can be very helpful to use as an image interpretation key while doing visual analysis.

How to link with Google Earth:

Open a viewer and bring in an image. Click on the Google Earth tab and then Connect to Google Earth. Google Earth will open in another window. In order to link the two windows, click on Match GE to View in Erdas, followed by Sync GE to View. Erdas and Google Earth will now move in sync.

Part 5: Resampling

Goal: Resample images based on analytical need

Resampling is the process of changing the size of pixels, either by Resample Up (reduce pixel size) or Resample Down (increase pixel size). The decision to resample up or down is determined by the type of analytical need.

In this section, two resampling methods will be used; (1) Nearest Neighbor and (2) Bilinear Interpolation.

1. Nearest Neighbor - Open an image in a viewer and open the Image Metadata. Be sure to take note of the pixel size. Close the Image Metadata window and click the Raster tab and then click Spatial > Resample Pixel Size. Inside the Resample window, select an input file and then under "Resample Method", accept the default Nearest Neighbor. Change the output cell size from 30x30 meters to 15x15 meters in both the XCell and YCell and be sure to check Square Cells to make sure that the output pixels are squared. Accept other default parameters and click OK.

 2. Bilinear Interpolation - Repeat the steps above, but instead of accepting the default, Nearest Neighbor, change the "Resample Method" to Bilnear Interpolation.

After creating image outputs with both resample methods, compare them side by side with the original image one at a time to notice how the pixel cells change.

Part 6: Image Mosaicking

Goal: Perform an Image Mosaic on a study area that traverses multiple satellite image scenes

In this section of the lab, two adjacent satellite scenes need to be mosaicked; path 25 row 29, and path 26 row 29. Both images were captured in May 1995 by the Landsat TM satellite.

There are two methods of mosaicking that will be covered; (1) Mosaic Express and (2) MosaicPro.

1. Mosaic Express - Both images to be mosaicked need to be opened in the same viewer. Before adding each image, make sure to click over to the Multiple tab in the Select Layer to Add window. Click the radio button for Multiple Images in Virtual Mosaic. Next, click the Raster Options tab and make sure that Background Transparent and Fit to Frame are checked. Click OK  to load the image to the viewer. Repeat for the second image.

Both images are now in the viewer and can be seen overlapping each other. The next step is to mosaic them together using Mosaic Express. Click on the Raster tab and then click Mosaic > Mosaic Express. On the Input tab, add the files/images to be mosaicked together, taking care to add the top image first, and then the bottom image second. Click Next and accept the default settings through each tab all the way to the Output tab. Choose a location for the output file and click Finish to run the model.

2. MosaicPro - Select both images to be mosaicked using the same steps as above. Now select Mosaic > MosaicPro. Click the Image Area Options tab and then the Compute Active Area button. Click OK on the Add Image dialog. Repeat the previous steps for the second image.

Now synchronize the radiometric properties in the overlapping area of the two images so that there will be a smooth color transition from one image to the other. In the MosaicPro window, click on the Color Corrections icon to open the Color Corrections dialog. Check the Use Histogram Matching option and click on Set. Select Overlap Areas as the Matching Method. Click OK in the Histogram Matching box and OK again in the Color Corrections dialog.

Click on Process > Run Mosaic in the MosaicPro window to run the model.

Part 7: Binary Change Detection (Image Differencing)

Goal: Estimate and map the brightness values of pixels that have changed in Eau Claire County and four other neighboring counties between August 1991 and August 2011

Section 1: Creating a Difference Image
The first step to mapping change over time is to create a difference image using a technique called Image Differencing. Open the 1991 image in one viewer and the 2011 image in another and sync the views. Click on the Raster tab and click Functions > Two Image Functions. The Two Input Operators window opens. Select the newer image as Input File #1 and the older image as Input File #2. Select the location for and name the output file. Under Output Options, change the Operator from "+" to "-". Click the Layer dropdown under Input File #1 and change from "All" to "4". Do the same for Input File #2. Click OK to run the model.

The next step is to determine the areas that have changed and the areas that have stayed the same by calculating the Change Threshold. Open the Image Metadata and record the Mean and Standard Deviation of the histogram to be used in the Change Threshold calculation (Mean + 1.5 Standard Deviation). Calculate the Change Threshold and add it to the Mean to get the upper limit of the change/no change threshold. Subtract the calculated Change Threshold from the mean to get the lower limit of the change/no change threshold.

Section 2: Mapping Change Pixels in Difference Image Using Spatial Modeler
Change over time can also be mapped by using the Spatial Modeler:

Open the Model Maker by clicking on the Toolbox tab and then Model Maker > Model Maker. Click the Raster Object icon and click inside the model panel to place the object. Place a second raster object next to the first. Next, place a Function Object into the model panel below the two raster objects. Finally, place a third Raster Object into the model panel below the Function Object. Connect all individual objects by using the Connector. 

Double click the Raster Objects and select the 2011 image as one input and the 1991 image as the input for the other Raster Object. Open the Function Object and enter in the equation ('2011.img' - '1991.img' + 127). Click the final Raster Object at the bottom of the model and set this up as the output. Run the model. Open the Image Metadata and view the Histogram to see the distribution and to calculate the change threshold at the upper tail of the histogram (Mean + 3 Standard Deviation).

 Next, create another model in Model Maker to actually show the pixels that changed, using the change/no change threshold that was calculated in the previous step. Create a new model with an Input Raster Object, Function Object, and Output Raster Object and connect them all. Select the difference image as the input raster. Open the Function and enter the definition as "EITHER 1 IF ( 'input file' > change/no change threshold) OR 0 OTHERWISE". Label the output binary image in the output raster and run the model.

The model has created an image that is difficult to read on it's own. Open ArcMap and overlay the Image Difference onto one of the other images of the area so that it can provide some context of the location of the changes. World Street Map is a good basemap to use in this situation and allows for additional analysis to be completed.

Results

The image outputs for each part of the lab can be found in this section.

Part 1: Image Subsetting (Creation of Area of Interest AOI) of Study Area

Goal: Subset satellite images of Eau Claire and Chippewa Falls area

Created a subset of Eau Claire and Chippewa Falls using the Inquire Box and creating a new output file, as seen in Image 1.1.

Image 1.1. Subset of Eau Claire and Chippewa Falls

Created a subset of Eau Claire and Chippewa Counties using an area of interest shape file. Many areas of interest are typically not perfect squares or rectangles, so this method is commonly used in those situations. See Image 1.2.

Image 1.2. Shape File subset of Eau Claire (Lower) and Chippewa (Upper) Counties

Part 2: Image Fusion

Goal: Use the 15m panchromatic image to ‘pan-sharpen’ the 30m reflective image

After doing a 'pan-sharpen' resolution merge, I made sure to take some time to analyze the differences between the two images. When I zoom in to the Eau Claire city limits, I am able to see more detail on the newly created ‘pan-sharpened’ image. The biggest resolution improvement that I have been able to locate is in the downtown area. The individual streets in downtown have become visible and more defined in the ‘pan-sharpened’ image, as seen in Image 2.1.

Image 2.1. Resolution comparison of Reflective Image (Left) and ‘Pan-Sharpened’ Image (Right)

There also appears to be more definition around the UWEC campus area, as well as numerous other locations within the image. One more thing that I noticed was the difference in reflection off of the water, especially in Dells Pond. The original reflective image has more blue/cyan colors over the water that match in color with the non-vegetative land surfaces throughout the rest of the image. This can be distracting to the eye and the newly ‘pan-sharpened’ image greatly reduces that.

Part 3: Simple Radiometric Enhancement Techniques

Goal: Perform preliminary radiometric enhancement techniques to enhance image spectral and radiometric quality

  After running the Haze Reduction process, the new output image appears to have much more clarity compared to the original image. The output image has lost the blue-ish haze that covered the entire original image and there is much more contrast between green vegetation (red) and non-green vegetation (blue) surfaces, making the image easier to look at. Notice the difference in Image 3.1; the image on the right is the output after the Haze Reduction process.
Image 3.1. Comparison of original reflective image (Left) and image after Haze Reduction process (Right)

Part 4: Linking Image Viewer to Google Earth

Goal: Link ERDAS Image Viewer to Google Earth

 An Image Interpretation Key assists in interpreting information present in an aerial or satellite image. Google Earth serves as a good Image Interpretation Key because it provides graphical and word descriptions of features within the study area. For Example, I was able to locate The Fire House on Google Earth (Image 4.1) within my study area of downtown Eau Claire.
Image 4.1. The Fire House in Downtown Eau Claire as seen in Google Earth

1.       Google Earth is an example of a Selective Image Interpretation Key because it uses images with supporting texts to identify features. Image 4.2 features an image of The Fire House and tells us that it is a sports bar.

 Image 4.2. Selective Key identification of The Fire House in Downtown Eau Claire

Part 5: Resampling

Goal: Resample images based on analytical need

1. Nearest Neighbor - The original image and the resampled image appear to still be the same after using the Nearest Neighbor method, even though the pixel size for the new resampled image is 15m x 15m. The Nearest Neighbor method transfers original data values without averaging them, and as a result, the new scaled up pixels in the resampled image keep the same data values, shown within the yellow outlines in Image 5.1. In other words, the four 15 x 15 pixels that make up one original 30 x 30 pixel have kept the same data values and therefore look the same.
Image 5.1. Comparison of original 30 x 30 pixel image (Left) and Nearest Neighbor resampled image 15 x 15 (Right)


2. Bilinear Interpolation - A difference can be seen when comparing the original image with the image that was resampled using the Bilinear Interpolation method. The Bilinear Interpolation method uses the brightness values of the four closest input pixels in a 2 x 2 window to calculate the pixel’s output value, and as a result, we can now physically see the increased pixel count in Image 5.2 and the color transitions between surface features appear smoother.

Image 5.2. Comparison of original 30 x 30 pixel image (Left) and Bilinear Interpolation resampled image 15 x 15 (Right)

Part 6: Image Mosaicking

Goal: Perform an Image Mosaic on a study area that traverses multiple satellite image scenes

The original images before any mosaic procedures can be seen in Image 6.1.

Image 6.1. Original pre-processed images of Eau Claire area

Using Mosaic Express to combine the original images, the resulting output, Image 6.2, still shows contrasting image layers that are especially noticeable along the boundaries. According to our instructions, eau_claire_1995p25r29.img was to be the first image added. In my opinion, the original images appear to be better blended than the Mosaic Express output.

Image 6.2. Mosaic Express output image of Eau Claire area

When comparing the outputs of Mosaic Express vs MosaicPro, it can be clearly seen in Image 6.3 that MosaicPro does a better job of blending the two input images together. MosaicPro requires more user input than Express and goes through an image smoothing and radiometric synchronization process to produce a visually pleasing output. The quality of the output using MosaicPro is fairly high, but it should still be pointed out that there is still one clearly visible boundary between the two original images.

Image 6.3. MosaicPro output image of Eau Claire area

Part 7: Binary Change Detection (Image Differencing)

Goal: Estimate and map the brightness values of pixels that have changed in Eau Claire County and four other neighboring counties between August 1991 and August 2011
Section 1: Creating a Difference Image
Estimate a cutoff point in the distribution using the equation: Change Threshold = Mean + 1.5 (Std. Dev)

Image Difference Output File:

Mean: 12.253                     Standard Deviation: 23.946                          Change Threshold = 48.172

Central Histogram Value: 24.199

Upper Threshold = 24.199 + 48.172 = 72.371         Lower Threshold = 24.199 – 48.172 = -23.973
Image 7.1. Change/No Change thresholds of pixel change distribution

Section 2: Mapping Change Pixels in Difference Image Using Spatial Modeler

Step 1. Calculate Change/No Change Threshold:

Change Threshold = Mean + 3 (Std. Dev)


Mean: 143.959                   Standard Deviation: 19.407                          

Change/No Change Threshold = 202.18
Image 7.2. Model used to calculate pixel differences between 1991 and 2011 images

Step 2: Develop model that shows the pixels that changed using change/no change threshold calculated in Step 1:

EITHER 1 IF ($n1_ec_91_11chg_b > 202.18) OR 0 OTHERWISE
Image 7.3. Model to show pixels that changed from 1991 to 2011

After overlaying the changed pixel output from the final function on top of an image of the area, the context of the changed areas can be seen and used for analysis (Image 7.4).
Image 7.4. Pixel Change Map of Eau Claire and surrounding Counties from 1991 to 2011

When looking at the spatial distribution of the areas that have changed, they would appear to be located close to urban areas like Menomonee and Chippewa Falls. There is also a noticeable change along the heavily traveled Hwy 53 corridor and also around he Chippewa River between Eau Claire and Durand. In order to better reference the location of the changes as seen in Image 7.5, I imported World Street Map to use as my basemap.
Image 7.5. World Street Map pixel change overlay

 Sources

Satellite images are from the USGS Earth Resources Observation and Science Center:

United States Geological Survey. Earth Resources Observation and Science Center. Retrieved on March 30, 2018.


Shapefile is from the Mastering ArcGIS Dataset:

Price, Maribeth (2014). Mastering ArcGIS 6th Edition Dataset. McGraw Hill. Retrieved March 30, 2018

Comments

Post a Comment

Popular posts from this blog

Remote Sensing - Lab 7: Photogrammetry

-
Image
Goal and Background The goal of this lab exercise is to become familiar with performing key photogrammetric tasks on aerial photographs and satellite images. Tasks include performing Stereoscopy and performing Orthorectification on satellite images. Completing this lab exercise will help to gain an understanding of the mathematics behind the calculation of photographic scales, measurement of areas and perimeters, and calculating relief displacement.  Lab 7 is split into three parts; (1) Scales, Measurements and Relief Displacements, (2) Stereoscopy, and (3) Orthorectification. Each of these parts are outlined in the Methods and Results sections of this blog post. Methods The methods used to obtain the results for each of the three parts of this lab exercise are outlined below. The data sources used for each part are listed in the Sources section of this blog post. Scales, Measurements and Relief Displacements Calculating Scale: If the actual distance between tw...
Read more

Remote Sensing - Lab 6: Geometric Correction

-
Image
Background and Objectives Prior to the extraction of biophysical and sociocultural information from satellite images, the analyst must perform some preprocessing activities. The following lab exercise is designed to introduce the concept of image preprocessing known as Geometric Correction . There are two major types of geometric correction that are normally performed; Image-to-Map Rectification and Image-to-Image Registration. Image-to-Map Rectification In this part of the lab exercise, the analyst will use a United States Geological Survey (USGS) 7.5 minute digital raster graphic (DRG) image of the Chicago Metropolitan Statistical Area and adjacent regions to correct a Landsat TM image of the same area. In order to accomplish this, ground control points (GCPs) must be collected from the USGS 7.5 minute DRG and used to rectify the Landsat TM image. Image-to-Image Registration In the second part of the lab exercise, the analyst will use an already corrected Landsat TM image ...
Read more

GIS I - Lab 2: Downloading GIS Data

-
Image
Goal and Objectives The goal of this lab exercise is to learn how to download and map data from the U.S. Census Bureau. This particular lab will focus on Wisconsin census data. Objectives: 1. Download 2010 Census data (total population) from the US Census Bureau 2. Download a shapefile of the 2010 Census boundaries from the US Census Bureau 3. Join the downloaded data to the Census shapefile 4. Map the data 5. Download and map a variable of your choice 6. Build a layout with both maps 7. Build a web map with one of the variables Instructions to recreate each of the objectives completed in this lab can be found in the Methods section of this post. Methods This section outlines the steps taken to complete each of the learning Objectives for this lab exercise. These procedures can be used in the future to recreate this lab, or can be used as supplemental material when performing similar tasks in the future. Objective 1: Download 2010 Census Data from t...
Read more