Miscellaneous Image Functions

Goal and Background:

The goal of this lab is for understanding to be gained regarding image processing, enhancement, study area delineation, mosaicking, and constructing processing models.

Methods:

Part 1: Image Subsetting

In part one we created subsets of images, using two different methods: the inquire box, and an area of interest.

Section 1: Subsetting with an Inquire box

In order to create a subset with the inquire box tool, I selected the "inquire box" tool from the raster tools. Next, I expanded the box so that it covered Eau Claire and Chippewa Counties, then clicked "apply". Next, I used the "Create Subset Image" tool from "Subset & Chip" to create a subset image from the inquire box.

Section 2: Subsetting with the use of an area of interest shapefile

In order to create a subset with a shapefile, I first added both the image and shapefile to a 2D viewer in Erdas Imagine. I then selected the two counties with my mouse, and clicked on "paste from selected object" to create an area of interest. Next, I saved this area of interest as an aoi file. After the file was done saving, I used the "Create Subset Image" tool from "Subset & Chip" using the aoi file instead of an inquire box to create the second subset image (Figure 2).

Part 2: Image Fusion

Image fusion is a process where two images are combined in order to optimize spatial resolution for image interpretation processes. One of such techniques is called "pan-sharpen", when a reflected image of coarser resolution is fused with a panchromatic (grey-scale) image of higher spatial resolution, to adjust the reflected image to the higher resolution. In order to do a simple pan-sharpen using Erdas Imagine, I first imported reflected and panchromatic images of Eau Claire taken in 2000 to a 2D viewer. Next, I selected the "Resolution Merge" tool from the "Pan Sharpen" dropdown menu in the "Raster" toolbar. I set the resolution merge tool to use: the panchromatic image as the high resolution input file, the reflected image as the multispectral input file, the "Multiplicative" method, and the "Nearest Neighbor" resampling technique (Figure 3).

Part 3: Simple radiometric enhancement techniques

Scattered electromagnetic radiation (EMR) causes haze to appear on remotely sensed images. In order to eliminate the haze, I used the "Haze Reduction" tool under the "Radiometric" tools dropdown menu of the "Raster" toolbar.

Part 4: Linking image viewer to Google Earth

When performing visual image interpretation, one method of obtaining more accurate classifications of ground features is by using ancillary datasets. One of such datasets is Google Earth. The "Connect to Google Earth" tool allows for the extent of the data frame and Google Earth to be synchronized.

Part 5: Resampling

In order to perform analysis between two rasters, they first need to have the same spatial resolution. Resampling is the method of changing the pixel sizes of a raster. Different types of resampling techniques are used for different purposes, as they all perform the same task in varying ways. In order to resample a Landsat image from 30m pixel sizes to 15m, I used the "Resample Pixel Size" tool under the "Spatial" tools dropdown menu of the "Raster" toolbar. I used two different resampling methods, the "Nearest Neighbor", and "Bilinear Interpolation".

Part 6: Image Mosaicking

When your area of interest falls between two separate images, they need to be mosaicked into one image before analysis can be done. There are numerous ways to do this, the easiest of which is called a virtual mosaic. Virtual mosaic requires no calculations, and is merely overlapping the two images in the same viewer. The other methods, however, join the two into one continuous image.

Section 1: Mosaic Express

Mosaic Express is a simplified version of the mosaicking process that requires minimal user input, and delivers results by a fully automated process. In order to perform Mosaic Express, I first added the two rasters to be mosaicked into a new 2D viewer with the raster I wanted to be in the front added first. Next, I selected "Mosaic Express" from the "Mosaic" tool dropdown menu of the "Raster" toolbar to perform the image mosaicking.

Section 2: MosaicPro

Mosaic Pro is a more advanced mosaicking method that requires more user input, but has substantially greater capabilities. In order to perform mosaic two images using MosaicPro, I first had to add two images to a new 2D viewer exactly as I did for the Mosaic Express. Next, I opened the "MosaicPro" tool from the "Mosaic" dropdown menu of the "Raster" toolbar. I used the "add images" dialog to add both of the images to the tool, with "Compute Active Area" set. I then used the "Color Corrections" dialog to set histogram matching for the overlap areas. After all of these parameters were set, I performed the image mosaicking.

Part 7: Binary Change detection (image differencing)

Binary change detection is a process that uses the change in pixel brightness values between two time periods to determine the extent of change that occurred between the two dates.

Section 1: Creating a difference image

In order to determine the extent of the change, a difference image is created. This is done using the "Two Input Operators" tool under "Two Image Functions" in the "Functions" drop-down menu of the "Raster" toolbar. The brightness value from the image taken in 1991 was subtracted from the brightness value of the 2011 image. After creating the difference image, I analyzed the histogram to determine the change threshold limits.

Section 2: Mapping change pixels in difference image using spatial modeler

I created a model to eliminate the negative values in the difference image, to easily show changes of features between the two dates. The model contained the 1991 and 2011 band 4 rasters, an equation in the form of a tool, and an output raster. The equation subtracted the 1991 brightness values from the 2011 image and added 127 to the resulting value to ensure all of the resulting values are positive. I analyzed the output image's histogram in order to determine the change threshold for the image was 202.18. I used this new information to create a second model, using the output image from the previous step, with a conditional statement to create an output raster with values of 1 if the amount of change was greater than the change threshold, or 0 if the amount wasn't greater than the change threshold.

Results:

Part 1: 
Section 1:

The inquire box generated a subset quickly, but it wasn't possible to see exactly where the boundaries of the counties were, so I ended up guessing (Figure 1).

Figure 1: Eau Claire and Chippewa Counties subset from an inquire box.

Section 2:

Using an area of interest to generate a subset provided a much more accurate subset image than the inquire box (Figure 2).

Figure 2: Eau Claire and Chippewa Counties subset with the use of an area of interest generated from a shape file.

Part 2: 

The pan sharpened image was much more suitable for image interpolation purposes, as it showed much more detail than the original.

Figure 3: The original reflected image to the left, with the pan sharpened image to the right.

Part 3:

The haze reduction tool reduced the effects of atmospheric scattering and shows the true reflectance values more accurately (Figure 4).

Figure 4: The original image is to the left, and the haze reduced image is to the right.

Part 4:

Linking the viewer with google earth allowed for image interpretation to be aided by ancillary data.

Figure 5: Landsat imagery is shown to the left, with corresponding google earth imagery to the right.

Part 5:

The nearest neighbor resampling method produced identical results to the original image. The bilinear interpolation resampling method produced substantially smoother, and more aesthetically pleasing results, but the actual reflectance values are less accurate as a result of averaging.

Figure 6: The nearest neighbor (left) doesn't average any values, causing it to be accurate but less aesthetically pleasing.
Bilinear interpolation (right) averages each cell's values by the values of the cells directly bordering it producing a more aesthetically pleasing image. The resulting image is reflectively inaccurate as a result of this averaging.

Part 6:

Before performing any of the image mosaicking functions, I created a virtual mosaic by stacking the images over one another (Figure 7).

Section 1:

The mosaic express created one cohesive image out of two separate images. The image didn't truly seem like one image, however, as there was a large color discrepancy between the two (Figure 7).

Figure 7: The virtual mosaic (left) isn't a true mosaic, as the two images aren't actually joined. Mosaic express (right) joined the two images, but didn't perform any other adjustments, leaving the resulting image rather jagged.

Section 2:

MosaicPro's histogram matching created a seamless image, where it is difficult to see where one image ends and the other begins (Figure 8).

Figure 8: The mosaic express (left) looks substantially more disjointed than the seamless image created by MosaicPro (right)

Part 7:

Section 1:

Image differencing created a histogram, which was used to calculate the upper and lower limits of the change - no change threshold (Figure 9).

Figure 9: The upper limit is 72.0544, and the lower limit is -23.2896.

Section 2:

The results of band 4 change detection between 1991 and 2011 were most dramatic in rural areas (Figures 10,11).

Figure 10: The change image (left), 2011 image (left-center), 1991 image (right-center), and 2014 Google Earth image (right).

Figure 11: The results show that the majority of NIR change between 1991 and 2011 happened in rural areas. 

Conclusion:

This lab provided a good introduction into many different image processing techniques used by remote sensing technicians. I anticipate using many of these techniques in the future, regardless of my career path.