Lab 4: Miscellaneous Image Functions

Goals and Objectives:
The main goal for this lab is to be able to delineate a study area from a satellite image, demonstrate how spatial resolution of images are optimized for viewing, introduce radiometric enhancement techniques in optical images, link satellite images to Google Earth to be used as additional images, explore image mosaicking, and expose students to binary change detection through graphical modeling. At the end of this lab, students will be able to demonstrate image reprocessing, enhancing images for interpretation, delineate any study area from a larger satellite image scene, mosaic multiple image scenes, and build a simple model for remote sensing analytics.
Methods
This lab ERDAS Imagine 2015, Google Earth and ArcMap were used in order to analyze information pertaining to the Eau Claire area. Image subsetting (creation of AOI) of a study area was analyzed by analyzing images of Eau Claire using the Raster tool in Erdas. The specific images were then subsetted with the use of an area of interest shape file to highlight to counties. By using the Raster and Subset & Chip tool, the output image is a subset of the original, subsequently creating an AOI. The AOI is Eau Claire and Chippewa Counties.



 Open two viewers of Eau Claire in 2000, one image is panchromatic. Using pan-sharpening and resolution merging, the nearest neighbor technique is used to pan-sharpen the 30 meter reflective image to 15 meters.

 Using simple radiometric enhancement techniques, haze reduction can be implemented through the haze reduction tool. Google Earth can be linked to Erdas using the Connect to Google Earth tool. Load an image of Eau Claire and connect to Google Earth and match GE to view to obtain the same spatial extent of the image. By synchronizing views, the same image can be viewed in Google Earth with certain advantages. Resampling was then utilized by viewing an image of Eau Claire. By using the Resample Pixel Size tool, the nearest neighbor method can be used to resample pixel size from 30x30 to 15x15. The process is repeated but with bilinear interpolation versus nearest neighbor. Next, image mosaicking is implemented with multiple images. Using Mosaic Express, two images are selected and run under default parameters. The following image is produced:

MosaicPro is then used on the same two images. By clicking the MosaicPro option, the active area can be manually computed to seam the two images together. Selecting the Use Histogram Matching option and selecting Overlap Areas allows for only the overlapped areas to be corrected, preserving the brightness values for the other areas of the images. Clicking on Process and then Run Mosaic produces this image:

Finally, creating a difference image using binary change detection is implemented by using two Eau Claire images. By clicking Two Image Functions under Functions in Raster, the two Eau Claire images can be filed under their respective input. The operator is changed from (+) to (-), and the layers for the images are reduced to four. Image differencing is then ran. By viewing the histogram in the image's metadata, the standard deviation and mean are observed to gather the upper and lower bands with the equation "mean + 1.5 x standard deviation", displayed below:

After this, the spatial modeler was used with Model Maker. Two raster objects were created to be input into a function. A third raster object is then placed and all the objects are connected using connectors. The two Eau Claire images are then used for the raster objects. The function used for the two images are "image1 - image2 + 127". Run the model to view the final image. The metadata is then observed for the image along with the histogram. In order to find the upper and lower bounds, the equation "mean + 3 x standard deviation" is used. Model Maker is then used to show the pixels that changed between 1991 and 2011 using the change/no change threshold, or upper and lower bounds found above. Input the differenced image as the input raster. Then run the function as Either IF OR with the change/no change threshold value. Run the model. The map is then displayed in Erdas, showing the pixels that changed between the two images. ArcMap is then used two more easily display these changes by overlaying the 1991 image as seen below:

Results
Through the use of the Subset & Chip tool, a subset of an image could be taken from a larger one in order to observe specific areas. In the pan-sharpened images, the finished product appears smoother and displays greater detail of areas such as roads and farming areas instead of harsher pixelated originals.
In haze reduction, images appear much darker and clearer than the original. Bodies of water are nearly black, and clouds are less apparent and are nearly imperceptible.
When linking Erdas to Google Earth, it can be used as a tool for association in image interpretation. When an image becomes too pixelated when zoomed in Erdas, the Google Earth view has a much sharper quality which can be used to identify buildings and other structures that otherwise couldn't be determined in Erdas.
Images were then resampled using the nearest neighbor method. There is very little difference in appearance between the resampled image and the original. This is because the brightness value of the closest input pixel for assigning values is the same as the output, effectively creating the same image. Bilinear interpolation, however, uses the brightness values of four of the closest input pixels in a 2x2 window to calculate the pixel's output value. This creates a sharper image when zoomed than the original.
Mosaic Express is then introduced. Through the use of Mosaic Express, the images are seamed together, but there is not a smooth color transition between the two at the boundaries. The top layer is clearly darker then the bottom and it is apparent where the two images intersect. MosaicPro, however, creates a nearly seamless transition between images. The colors also match between both images more closely than with Mosaic Express. The MosaicPro image is more accurate because only the areas that overlapped were interfered with, instead of the entirety of both of the images.
After mosaicking is implemented, image differencing is used to view two Eau Claire images. Using the metadata and histogram of the final image, the upper and lower band were determined with the given equation. The lower band resulted in -24.47 and the upper 71.8.
Finally image differencing is implemented to show the difference between Eau Claire in 1991 and 2011. The spatial distribution areas in red above are close to urban centers. It shows the activity and growth of these areas, showing change in the land over the last 20 years. However, these areas are fairly scattered with only a few central changing areas. This could be a result of deforestation, changes in crops, or expansion of urban areas.




Source: Cyril Wilson, Geog 338, Fall 2015