Wednesday, April 20, 2016

Geometric Correction

Goals and Introduction

The goal of this lab was to learn a few ways to correct an image by looking at a reference image. We specifically learned Image to map rectification and image to image registration. To do this Ground Control Points (GCP's) were added to the images to correct them. This process is explained more below.

Methods

This lab started by creating a lab 6 folder within our personal folder where we would save any and all outputs. To start the image to map rectification we first added the Chicago_drg.img in one viewer and the Chicago_2000.img image in another viewer side by side. Next I clicked on Multispectral to activate the raster tools and then clicked on control points. This opened the "Set Geometric Model" where I then clicked Polynomial. The default was then accepted on the Image Layer. The reference image Chicago_drg.img was then added where I clicked okay for the Reference Map Information dialog. The Chicago_2000.img was also added to be corrected. Since we performed a 1st order polynomial we only needed to add three GCP's but added four to make sure the image would correctly rectify.

Part 1

I clicked on the GCP tool and added a GCP on the input image (Chicago_2000.img) and then added a GCP in the same exact place on the reference image (Chicago_drg.img). I did this exact same thing for the next two GCP's. This changed the status of the image from "model has no solution" to "model solution is current." The fourth GCP was added to only one image and it automatically placed the corresponding GCP in the other image.

Next the GCP's were evaluated, this is done by Root Mean Square (RMS) Error. The ideal RMS error should be 0.5 or below. Currently at this time the RMS error was well above 2.0. To fix this, each ground control point was zoomed into and then carefully moved around while watching the RMS Error. The goal was to get the total RMS Error value below 2.0 (since we were beginners for this lab). This was done for each GCP until the total RMS error was below 2.0.

Now the geometric correction is ready to be done on the image. The "compute transformation matrix" has already been computed from the Ground Control Points. I then clicked the "Display Resample Image Dialog" button and the image was saved in my lab 6 folder as Chicago_2000gcr.img. All parameters were left as they were and the geometric correction output ran. I did not save the points on either image as we were told to not to save them.

Part 2

This part of the lab was image to image registration and was also done in Erdas Imagine. I worked with the sierra_leone_east1991.img which had serious geometric distortion. We looked at the distortion by overlaying another image named sl_reference_image.img and then activated the swipe tool to compare images and just see how much distortion there really was. After looking at the distortion I began to fix it by bringing in the distorted image (sierra_leone_east1991.img) and in the second viewer I brought in the reference image (sl_reference_image.img).

Once again I clicked on Multispectral to activate the raster tools and then clicked Control Points. I then clicked on polynomial in the set geometric model and then OK on the Collect Reference Points From in the GCP Tool Reference Setup. Next I added the sl_reference_image.img and clicked OK on the Reference Map Information and then on the Polynomial Model Properties I changed the Polynomial order to 3 and clicked close.

Now that it was set up for the GCP's I clicked the "Create GCP" tool. I added a GCP first to the input image (sierra_leone_east1991.img) and then added a GCP to the same exact place on the reference image (sl_reference_image.img). I continued this process for points 2 through 9. After adding 9 points the model still stated "Model has no solution" so three more points were added. I then moved and repositioned the Ground Control Points until the RMS error was less than 1.0.

Next I clicked the "Display Resample Image Dialog" button on the Multipoint Geometric Correction tools bar. The output image was saved as sl_east_gcc.img in my lab 6 folder. I then changed the resample method in the resample image window to bilinear interpolation, the defaults were accepted and the geometric correction output was executed. No points were saved for either of the images.

Results

Figure 1: First image corrected with 4 Ground Control Points
Figure 2: Showing the Total RMS error is 1.33 (less than 2.0)
Figure 3: Second image corrected with 12 Ground Control Points
Figure 4: Showing the total RMS error of this image was 0.9099 (less than 1.0)

No comments:

Post a Comment