Geometric Correction

(also known as georeferencing and image rectification)

  • Before you begin, you will need either a previously georeferenced image, a GIS database (e.g. road network), or a good map covering the area in your image.  This georeferenced image / GIS database / map must have the coordinate system (i.e., latitude-longitude, UTM, etc.) that you wish to use for your new image clearly marked on it.
  • Geomatica manages the geometric correction process through the OrthoEngine application.  Start OrthoEngine from the Geomatica toolbar:

 

  • When you first start OrthoEngine most of the functions will be dim and unselectable until you start an OrthoEngine project.
    • If you are just starting a geometric correction process for the first time, select New from the File menu.
    • If you are returning to a previously saved project, you can Open... it.
  • When the Project Information panel appears, fill in the following:
    • Give your project a Filename.  It makes sense to give the project the same name as the image you will be correcting (the system assigns different file extensions to the project and image files to keep them separate).
    • Provide a short Name for your project that will be shown in lists of similar project files.
    • You can also enter a longer Description of what your project is about.
    • For a typical geometric correction, select a Polynomial Math Modelling Method.
    • Click OK.
  • You specify the Earth coordinate system you will be registering your image to in the Set Projection panel.  
    • Specfiy the coordinate system for your corrected image by selecting the appropriate item from the Metre pull-down menu.
      • This is the projection that is used in your georeferenced image / GIS database / map.
      • UTM is a good choice for most Canadian areas.

 

  • Select the appropriate Datum for your projection.  This should match the information printed on the base map that you are using.
    • Most Canadian topographic maps published before 1985 use NAD27.
    • Most Canadian topographic maps published after 1985 use NAD83.
  • Click Accept.

 

  • Select the appropriate UTM Zone and Row for the area of your image.

 

  • Now you can specify the size of the pixels in the corrected image that will be created.  By convention, this is usually an even divisor of 100 that is the same, or slightly smaller than the resolution of the uncorrected image.  (Note: although the output pixel spacing may be different than the original data, this does not mean that the spatial resolution of the image has changed). Some common pixel sizes are:
Sensor Spatial Resolution Output Pixel Spacing
Landsat TM/ETM 30 m 25 m
Landsat ETM Pan 15 m 10 m
SPOT HRV 20 m 20 m
SPOT PAN 10 m 10 m
IKONOS 4 m or 1 m 1 m
QuickBird 0.6 m 0.5 m
  • Click on Set GCP Projection Based on Output Projection.
  • Click OK to close the Set Projection panel.
  • This is a good time to save your work so far.  Select Save from the File menu.

 

  • Now you are ready to move on to the next stage of the process. Select GCP Collection from the Processing Step pull-down menu. Note how the toolbar buttons change to match the processing step you are in.

 

  • Select the image that you want to correct:
    • Click on the Open a new or existing image button.

 

  • The Open Image panel lists images that you have previously worked on in this OrthoEngine project.  If you are creating a new project, this list will be empty.  Click on New Image.
  • Select the image file that you want to geometrically correct from the Database File Selection window.

 

  • After you have specified an image file, it will appear in the list in the Open Image panel.  
    • Click on the filename in the panel to highlight it.
    • Click on Quick Open & Close.
  • The image will appear in a new Viewer window.
    • The image viewer is split into 3 panes: 
      • The largest, right-hand pane shows a portion of your image at full resolution (1 screen pixel = 1 image pixel).
      • The top-left pane shows an enlargement of the area surrounding the current cursor location.  You can use the zoom tools to adjust the magnification in this pane.
      • The bottom-left pane shows an overview of your image.  You can view different sections of your image in the full-resolution pane by dragging the red box around in the overview pane.

 

  • You can also adjust how your uncorrected image is displayed by clicking on the RGB Mapper and/or Enhancements buttons.

 

  • Now you are ready to begin collecting ground control points (GCPs).  Click on the  Collect GCPs Manually button.
 

 

  • The objective of collecting GCPs is to accurately position the cursor over a location on the image that you can identify from a map.  The method you use to collect control points varies slightly, depending on your georeferenced source:  a map, another image, or a vector file.

    Select your desired GCP collection method from the GCP Collection Panel::

     

Manual entry

  • Using the pan and zoom tools, positioning the cursor exactly over the location in your image that you wish to use for control.  Good choices are road intersections in built-up areas or sharp river bends in natural environments.

  • Click on Use as GCP.  This will transfer the image coordinates (pixel and line values) for your point into the GCP Collection pane.
  • Now read the coordinates for this point from your map as accurately and with as most precision as possible.  Enter these coordinates into the appropriate section of the GCP Collection window.
  • Click Accept.

Geocoded image

  • In the Auxillary Information section near the bottom of the GCP Collection window, click on the Load button.

  • Browse to the geocoded image you are going to use, select the bands for display, and click Load and Close.

  • Using the pan and zoom tools, positioning the cursor exactly over the location in your uncorrected image that you wish to use for control.  Good choices are road intersections in built-up areas or sharp river bends in natural environments.

  • Click on Use as GCP.  This will transfer the image coordinates (pixel and line values) for your point into the GCP Collection pane.
  • Now position the cursor exactly over the identical location in your geocoded image.
  • Click on Use as GCP.  This will transfer the image coordinates (pixel and line values) for your point into the GCP Collection pane.
  • Click Accept.

 

Geocoded vectors

  • In the Auxillary Information section near the bottom of the GCP Collection window, click on the Load button.

  • Browse to the vector file you are going to use.  (Tip:  Select the .shp file if you are using ArcView Shapefiles). Click OK.

  • Highlight the Vector Segment you want to use and then click Load & Close.

  • Using the pan and zoom tools, positioning the cursor exactly over the location in your image that you wish to use for control.  Good choices are road intersections in built-up areas or sharp river bends in natural environments.

  • Click on Use as GCP.  This will transfer the image coordinates (pixel and line values) for your point into the GCP Collection pane.
  • Now position the cursor exactly over the identical location in your vector image.
  • Click on Use as GCP.  This will transfer the georeferenced coordinates for your point into the GCP Collection pane.
  • Click Accept.

 

  • Repeat this process until you have about 10-15 GCPs scattered all around your image.  Try to make sure all areas of your image are represented - i.e., near the corners and near the middle.
  • Before you proceed to the next step, change the Polynomial Order (found at the bottom of the GCP Collection window) to 2.  This is a reasonable order for correcting satellite imagery to ground coordinates.  Other orders are used for different types of image corrections.
  • Based on the points that you have given it, the program has calculated a polynomial transform (a mathematical equation) that it can use to convert your data from image coordinates to your georeferenced map coordinates.  There won't be an exact match for each point because a variety of imprecisions in the way the satellite acquires the image, the way maps are made, and in the method you are using to enter GCPs.  The program does its best to try and minimize the errors across all of your GCPs.  To give you an idea of how accurate the conversion process will be, the program uses the polynomial transform to calculate the proposed location for each of your points.  The difference between the proposed location and the actual values you entered is called the residual.  The residuals for each GCP are listed in the GCP Collection window:  Res X refers to the residual east-west difference between the calculated location and the value you entered.  Res Y is the north-south residual.  The values in the Residual column are the combined x- and y- residuals.
  • To evaluate the accuracy of the current polynomial, refer back to the GCP Collection window:
    • Examine the RMS (root mean squared) error estimates.  The X RMS is the average residual in the east-west direction.  The Y RMS  is the average north-south error.  the RMS is the overall error estimate for this set of points.  The objective is to arrive at an overall RMS error of less than 1 pixel.  

     

    • The GCPs are listed in the window in order of decreasing residual error, i.e., from worst to best.
    • You should re-examine those GCPs with high residuals for errors in data input.  Go back to the image to make sure you positioned the cursor as accurately as you could.  Re-read the map coordinates to verify what you have entered.
    • To edit a GCP value, click on it and you will be able to enter new coordinates in the upper section of the GCP Collection window.
    • Sometimes a GCP appears to be accurately positioned, but still has unacceptably high residuals.  There are 3 actions you can take in this situation:
      • add a new GCP near the poorly registered one to see if this improves the RMS error;
      • temporarily disable this point to see if this improves the overall RMS error (the error for the individual point may worsen, but since you have disabled it this is of no consequence);
        • select the point from the list
        • click on the button labelled GCP near the very top of the GCP Collection window and select Check from the pull-down menu
        • click Accept
        • note that the Type changes from GCP to Check in the GCP list
        • to re-enable the point, select it and then change its type back to GCP
      • permanently delete this point 
        • select the point from the list
        • click on the Delete button
        • be sure to add a new point for every point you delete.

       

  • Once you are satisfied with your overall RMS error, you can Close the GCP Collection window.

 

  • This is a good time to save your work so far.  Select Save from the File menu.
 
  • Now you are ready to move on to the final stage of the process. Select Geometric Correction from the Processing Step pull-down menu. Note how the toolbar buttons change to match the processing step you are in.
    • If you are re-starting a new correction based on previously saved work, you can open the image file again with the Open Image toolbar button.
    • The Define Clip Region button is useful if you wish to only correct a sub-region from your original image.  This is not normally required.
    • To proceed, click on the Schedule Geometric Correction toolbar button.
  • Your uncorrected image should appear in the Available Images list in the Geometric Corrected Image Production window.
    • Click on your image's name and then on the arrow button to put it into the Images to Process list.

  • Note that in the Corrected Image pane, much information has already been filled in for you.  The program is proposing to create a new file for your geometrically corrected images.  By default this file will:
    • be given the same name as the uncorrected image file, preceded with a lower-case "o", to signify that it has been processed by the OrthoEngine;
    • contain corrected versions of all of the channels in the uncorrected file; and
    • have the minimum and maximum bounding coordinates listed.
  • You may change any of these options, if you wish.

  • Under Processing Options select your desired Resampling algorithm
    • Nearest Neighbour is required for classified data
    • Cubic Convolution generally performs the best for continuous (non-classified) data.
  • When you are satisfied with your selections, click Correct Images.

  • After the geometric correction has completed, you can:
    •  Close the Geometric Corrected Image Production window;
    • request a Residual Report by selecting Reports from the Processing Step pull-down menu;
    • do a final Save of your project (File menu); and
    • Exit OrthoEngine (File menu).
  • Have a look at your new image in a Focus window.  Its always a good idea to verify your correction by comparing a few georeferenced coordinates on the image against a map. 

 



Geomatica Visual Guide


Revised: 2010-10-15
Copyright Joseph M. Piwowar