Lab 6: Geometric Correction

 

 

For all activiaties done in this lab, the images used were from The United States Geological Survey 7.5 minute digital raster graphic data collection.

Part1: Image-to-map rectification

In the first part of the lab, students worked with image-to-map rectification operations. In this case, the image and map in use were covering Chicago and surrounding areas near the Wisconsin and Illinois border.  
students conducted a first order polynomial special interpolation technique. to do this students placed ground control points (GCP's) on the same locations on the image (distorted Image) and the map (reference source).  The computer than uses these GCP's in algorithms that adjust the image to the points on the map, creating an image that is now has accurate geographic properties, closer to their real world location. The number GCP's needed for a given operation depends on the degree of the polynomial operation being done. Since the adjustment is only a minor first order polynomial (linear adjustment) its only necessary to use 3.  With that being said, its advised by the lab to use 4 for the sake of maximized accuracy.  Since the adjustment being made are relatively minor, it is appropriate to just use a nearest neighbor resampling method.

I was able to reach RMS error values lower than the requested 2 % for all four of my GCS points. One should always strive to have the lowest values possible for RMS error in order to ensure an accurate image .

Part 2: Image-to-image rectification

This part of the lab covered a part of Sierra Leone, Africa, and required that I do similar steps as I used above, but this time the components are both images.  There is one image that is geometrically accurate and one that is not.  This can easily bee seen by simply overlaying the two images, and using the swipe tool to see how the features are off from there actual location. Due to the high level of distortion, the operation used was a 3rd order polynomial spatial interpolation. This required that we place a minimum of 10 GCP's on both the reference and distorted image. For the sake of increased accuracy I used 2 extra GCP's, using a total of 12 to correct the distortion.  
Due to the high level of correction being done, the resampling method I used was the Bilinear Method.  This was the selected method because there was more pixel redistribution due to the high level of the original distortion.   

The image produced was still slightly distorted, highlighting the fact that getting the lowest value of RMS error is critical.  even the lab suggested that I get lower than 1.0% for all the points, its advised to get that value as low as possible for the sake of accuracy.

 



PART 1: Image to map rectification: map on the right is the reference source, the image on the left is the distorted image.  The highlighted blue column is the RMS error.

 

 

 

 

PART 2: Image to image rectification: map on left is the distorted image, map on the right is the geometrically accurate Image.

PART 2: This is the corrected image being laid over the already corrected image, Its nearly perfect wit still some distortion visible at the corners.

 

 

 

 

 

Lab 5: Image mosaic and miscellaneous image functions 2

This lab was given to us in order to introduce us to various operations in ERDAS imagine like: image mosaicking, spatial/spectral enhancement, band ratio, and binary change detection.  Each of these operations, in one way or another, allows you to manipulate an images spectral or spatial qualities in order to make better interpretations about what important features or details an image posses.

Part 1: Image mosaicking

Image mosaicking is the process of combining two or more images into one, in order to increase the visible area of study.  In this lab, we used two different mosaic operations: Mosaicexpress and MosaicPro.





mosaic pro

mosaic express

As you can tell by the images, mosaic pro offers a much smoother and more aesthetic image.  This is because through this operation one has much more user input meaning students have more options in terms of ways the pictures can be synced into one. In this case, we did histogram matching.  Mosaic express is a much simpler operation that only requests an input file and the name of an output file.



 Part 2: Band ratioing.


In this part of the lab, students used the NDVI index  (normalized difference vegetation index) which can be summarized as (NIR - Red bands / NIR + Red bands) in order to highlight areas of high vegetation and low vegetation.

the areas that are not white indicate the patches of the earth where vegetation has been removed

Part 3: Spatial and Spectral Image enhancement

In the first part of this section, students used spatial enhancement techniques to alter the frequency of an image. frequency is defined as the rate at which brightness values change over a given space.

the image on the left is the original. the image on the right is the image after the 5x5 low pass operation.

The operation conducted was a 5x5 lowpass spatial enhancement.  This operation was appropriate due to the high frequency of the original image, which made the image a very salt and peppery look to it.  conducting the lowpass operation made the image much smother and more consistently toned.



The image on the left is the original. The image on the right is the image after a 5x5 low pass operation.

 

Another filter that students were asked to use was Laplacian filter.  In this case we were instructed to use a 3x3 Lablacian edge detection operation.  The resulting image provided a more neutral appearance of the colors.  The filters intent is to increase the contrast at areas where there is transition.






The image on the left is the original, the image on the right is the image after the 3x3 Laplacian edge detection.


Section 2: spectral enhancements


In this portion of the lab students conducted various levels of spectral enhancements based of off the type of histogram that the original image had.  The operations used were min/max contrast stretches, which are most appropriate for images with low contrast (Gaussian or near Gaussian histograms).  The other operation we used was  a Piecewise stretch, which is more appropriate for histograms that have wider ranges of pixel brightness (none Gaussian histograms)







min/max contrast stretch. The lowest value of the histogram is stretched to 0 and the max value of the histogram is stretched to 273



piecewise stretch. A linear enhancement technique employed on various parts of the histogram, based off of the portion the user would like to enhance.  

The third technique of spectral enhancement we used was histogram equalization. This is a nonlinear method where pixels at the peak of the histogram are stretched, adding contrast.  At the same time, pixels near the end of the histogram are clustered, lowering contrast.  Overall though, the contrast of the image is increased and the histogram is flattened out.




the image on the right is the original. the image on the left is the same area after a histogram equalization operation.



 

 

 

 

Part : Binary change detection.

 

 

In this portion of the lab. students analyzed the pixel differentiations between two different images, taken years apart, of the same area.  Through a binary change detection, we could combine the two images into one.  This is done through a model of subtraction where you minus the 1991 image from the 2011 image.  Even with the resulting image, its nearly imposable to tell where the pixels are different (and there fore land use/land type is different). in order to tell where change had occurred we uploaded the image to ARC map.  on this program we overlaid the 1991 image with the image indicating the changed areas and were thus able to produce this map: 

 



 

 

 

 

 

 

 

All data and images used are from ERDAS imagine.