Lab 4: Introduction to Image Processing

Introduction:

Properly processing an image is important for visual interpretation in remote sensing.  Some processes that are important in processing an image are delineating a study area from a larger remotely sensed image, optimizing images for visual interpretation, using radiometric enhancement techniques, linking satellite imagery to Google Earth in order to use Google Earth as a key in image interpretation, and resampling satellite imagery using different methods such as nearest neighbors and bilateral interpolation.  These processes will be completed and detailed in this technical report.


Methods:

Image Subsetting:
Subsetting an image to better focus on a specific area can be done in two general ways.  One is through the creation of a rectangular box within a satellite image scene through the use of an Inquire Box.  This is done simply by clicking on the inquire box function and insuring it is surrounding the desired area to be analyzed.  The other, typically more useful method is by delineating an area of interest using a shapefile of the area of interest.  This method is typically more useful because the desired subset area usually isn't shaped like a rectangle and using a shapefile allows a more specific area of interest to be established.  In this lab a remotely sensed image which originally included a large portion of the western half of Wisconsin and stretched to the St. Paul, Minnesota was able to be subsetted to be focused over Eau Claire and Chippewa Counties using a shapefile of the two counties
(Figure 1).

This is an image subset of a larger original image.  This includes Eau Claire and Chippewa Counties.  The area of interest was delineated using a shapefile of Eau Claire and Chippewa Counties.  As it can be seen, this image subset is not rectangular.  If the Inquire Box method were used to subset this image it would include areas outside of the counties as it would be required to have a rectangular subset. (Figure 1)

Optimization of Spatial Resolution of Images:

Pan sharpening involves using a panchromatic band of an image, which is typically a higher spatial resolution, and using it to increase the resolution of a reflective image.  In this lab an image of Eau Claire and Chippewa Counties with a spatial resolution of 30x30 was able to be combined with a panchromatic image with a spatial resolution of 15x15.  This allowed the image to be pan sharpened.  It used the Pan Sharpen tool underneath the Raster tool bar in ERDAS Imagine.  This pan sharpened image appeared sharper with more contrast and clearer colors.  It was also easier to tell objects in the image apart thanks to the clearer imagery.

Radiometric Enhancement Techniques:
Haze reduction is a key way to enhance the resolution of an image.  Some images may appear cloudy and whited out; this is due to the large amount of haze in the imagery.  Thankfully, there are methods to reduce this haze.  Under the Raster toolbar and Radiometric, Haze Reduction can be found.  Inputting an image into the tool will help clear it up.  This process was done using an aerial image of the Eau Claire area.  Figure 2 shows how the Haze Reduction tool greatly improved the quality of a portion of the inputted image.

The image on the left is the input image, while the image on the right has been run through the Haze Reduction tool.  This haze reduction has eliminated the white cloud-like portions of the image, though some shadow-like portions remain.  The image as a whole is clearer with greater contrast and deeper colors than the input image. (Figure 2)

Linking Images to Google Earth:

Images in ERDAS Imagine can be linked to Google Earth.  That is to say, the same area can be viewed on an ERDAS image and in Google Earth at the same time.  This takes advantage of the high resolution Google satellites and allows Google's data to be used as a key to aid in image interpretation.  It's as simple as going to the Google Earth toolbar, clicking on "Connect to Google Earth" and "Match GE to View" (Figure 3).  The view in Google Earth will now match the view of the image in ERDAS.  From here the images can be synced to insure they remain in the same spatial context (Figure 4).

This image shows two views that have been matched.  The one on the right is in ERDAS and is an image of the Eau Claire area.  The image on the right is the same exact area as it has been matched to the image, though it is in Google Earth.  From here Google Earth can be used as a key to aid in image interpretation as many features are labeled in Google Earth and the image resolution is higher. (Figure 3)

This is a part of the Google Earth toolbar in ERDAS.  "Connect to Google Earch" simply brings up Google Earth along side ERDAS.  "Match GE to View" matches the image in ERDAS with that in Google Earth spatially.  "Sync GE to View" then insures that when the view in ERDAS is altered, it is similarly altered in Google Earth. (Figure 4)

Resampling Satellite Imagery:
Resampling is the process of changing the pixel size of an image.  It can be done to either reduce or increase the size of the pixels.  In this case it will be done to reduce the pixel size.  There are two common methods to resample an image: nearest neighbors and bilateral interpolation.  Starting with an image of the Eau Claire area, both nearest neighbors and bilateral interpolation were run in order to change the pixel size from 30x30 to 20x20 (Figure 5).  The differences can be seen in Figure 6 and Figure 7.  Though it's not readily apparent, there are some differences.

This image shows the how the pixel size has been altered.  The original image's metadata is on the right and shows a pixel size of 30x30.  The middle and left images have been resampled and have a pixel size of 20x20. (Figure 5)

This shows the difference between an image that has been resampled to a smaller pixel size using bilateral interpolation and the original image.  Upon close inspection the resampled image appears to have cloudier borders between objects but has noticeably less pixelization.  This appears to be due to the pixels blending together in a way. (Figure 6)

This image shows the difference between an image that has been resampled to a smaller pixel size using nearest neighbors and the original image.  The resampled image actually appears to be slightly more pixelated after using nearest neighbors. (Figure 7)

Conclusion:

Being able to process an image is key in using remote sensing effectively to answer a question or solve a problem.  Having a proper image is key in order to do work well.  Sometimes the image may be to hazy or may not be focused enough on a certain area.  This lab has taught the proper ways to go about preparing an image using image processing; whether it be resampling, haze reduction, or simply using Google Earth as a key to aid in interpretation.