Georegistration

Georegistration is the process of adjusting one drawing or image to the geographic location of a "known good" reference drawing, image, surface or map. For brevity, this topic and other georegistration topics use images as examples. However, the same procedures apply when georegistering drawings or surfaces .

 

·      The drawing, image, surface or map being used as a reference is called the reference component.

·      The drawing, image or surface being adjusted is called the target component.

 

images\ill_georegister_02.gif

 

Control points are used georegister images, surfaces, drawings and labels components. During georegistration the target image will be re-projected to match the reference drawing using the control points as a guide. The image above shows conceptually how control points in a reference drawing are matched up with equivalent control points in a target image.

 

Control points are matched by their name. For example, the spot marked by a control point named "ControlPoint2" in the drawing will be matched to the location in the image marked by a control point that is also named "ControlPoint2." Control points may be called by any names so long as the names used for control points within the reference component or target component correspond to the same location. For example, we could use the name "San Mateo Bridge East" for control points in both the reference component and the target component. These two points with the same name will be matched during the georegistration process. The choice of names is up to you. Some users prefer long, descriptive names while others prefer short names.

 

Any extra control points within either the reference component or the target component will be ignored. If there is a "ControlPoint18" in the target component but no control point of that name in the reference component it will be ignored.

 

We use the Control Points pane to add new control points to a drawing, image or map. Control points may be added to maps so that they may be used as a reference component, but maps are not georegisterable as a target component since they are made up of images or drawings in layers.

 

To Georegister a Target Component to a Reference Component

 

1. Place as many control points as is possible in the target image or drawing to be registered. Control points should be as evenly distributed as possible.

2. Place control points in the reference component with names matching the equivalent control points in the target.

3. Click on the image or drawing to be registered to make it the active window and choose Register in the Control Points pane. Manifold will present a dialog asking which component to use as the reference component. Only components with matching control point names will be listed.

4. Chose the reference component desired. Manifold will match those control points in the reference component to the target component by name and will re-project the target component so that its control points are in the same locations as the reference component.

 

Control Points

 

images\dlg_control_points.gif

 

The Control Points pane always shows a list of control points defined for the current image or drawing together with their coordinates. When we click on a different window the Control Points pane will switch context to show any control points defined for the active window.

 

Control points may be added by clicking with the mouse into drawings, images or maps or they may be added by manual entry of X / Latitude and Y / Longitude coordinates. See the Control Points pane topic for details.

 

The Control Points pane is also used to rename control points or to change their location by changing their coordinates. We can click into the Name field to change the name from the defaults, if desired. To change the coordinates of a control point we click into their X / Longitude or Y / Latitude fields.

 

Tech Tip

 

Uncheck the Autoscroll window on edit or selection operations option in Tools - Options when working with control points. Because using the New Control Point tool activates the mouse as soon as it leaves the Control Points pane, if this option is not deactivated the target window will begin to autoscroll as the mouse enters it.

 

Georegistering Images

 

The most frequent use of georegistration is to georegister images. Images, such as scanned paper maps, are often imported with the intent of tracing them to create new, custom drawings and maps. The first step in the tracing process is to georegister the image into the correct geographic location.

 

If the image includes features that also exist in a known good drawing we can use the drawing as a reference to georegister the image. For example, we may have scanned a paper map that shows property lines within a city boundary. We might already have a drawing that shows the city boundary but not the property lines. We can use the common feature of the city boundary to locate control points in both the reference drawing and the scanned image target. This procedure is discussed in the Georegistering an Image to a Drawing topic.

 

At times we need to georegister an image where we have no pre-existing drawing that has features in common with the image. To georegister such an image we need to identify features within the image for which we know the precise latitude and longitude coordinates. We can then use these features as control points based on the latitude and longitude coordinates they are supposed to have. For example, perhaps we have a small-scale paper map showing trails through a park where the park does not appear in any existing GIS drawing that we have. We would like to scan the map and georegister it so we can trace it. We can physically go out to different locations shown in the map and measure their precise coordinates with a handheld GPS device and then use these locations as control points. This technique is discussed in the Georegistering an Image to Known Coordinates topic.

 

Georegistering an image re-projects that image. Re-projecting an image invariably involves re-sampling the image by interpolating the image to a greater or lesser number of pixels. Images that are imported into Manifold from geographically-mute formats such as .jpeg or .gif do not have any scale associated with them. Georegistering such images into a geographic context will be the first time that the "size" of a pixel matters. By default, Manifold will attempt to preserve approximately the same number of pixels in height and width in the image. This can be adjusted if desired by changing the Size parameters in the georegistration dialog.

 

Georegistering Drawings

 

Drawings may also be georegistered. When we insert a new, blank drawing into a project it appears as a drawing in orthographic projection centered on the 0,0 origin of the Prime Meridian with the Equator. We can work with such drawings as if Manifold were a CAD editor to create blueprints, factory plans and other drawings where we do not care if they have or do not have a geographic context. We can also import pure CAD drawings from editors such as AutoCAD where the drawing has no geographic context.

 

We may wish to use such "CAD" drawings within geographic maps. For example, we may wish to place a CAD drawing of a real estate development within a geographic map of a town. To do this we can georegister the drawing to the desired location as is shown in the Coordinates Tutorial topic.

 

Georegistering Surfaces

 

Surfaces representing terrain elevation rarely need to be georegistered since most formats used in GIS to convey digital elevation can be imported as georegistered images. On occasion one encounters a format from which surfaces are imported that must be georegistered.

 

Georegistration Dialog

 

images\btn_register.gif To georegister the active image or drawing, click on the Register button in the Control Points pane.

 

Note: Some options will be enabled only for certain methods or when sufficient control points have been defined.

 

Reference

The name of the drawing or image to be used as a source of control points to be used as reference points.

Method

One of the following algorithmic methods:

Affine (triangulation) - Also called geometric transforms, affine methods can georegister a target component to a reference component using fewer control points than required by the Numeric method. However, affine registration requires care in placement of control points.

Affine (scale, shift, rotate) - A fast method that is algorithmically equivalent to using Numeric with an Order of 1. This is the default method since it works with any number of control points.

Simple (scale, shift) - Match components using XY translation and re-scaling only. Works with any number of control points.

Numeric (polynomial) - Numeric matching uses numerical computation methods to transform one component to match another. A large number of control points are required for good matching but the method is algorithmically simple and fast. Available only when eight or more control points have been defined, thus allowing an order of 2 or greater.

Note: Only those methods usable with the number of control points you have defined will be displayed. If you don't see a method, add more control points.

Order

The level of mathematical sophistication applied. Higher orders result in better matches but take more time and more control points. In numeric methods the highest order exponent used in the polynomial equations generated to transform the coordinate system of the target component. Enabled only for the Numeric method.

Modify Coordinate System

Enabled when the Simple method is selected. If checked (default) the coordinate system (projection) of the component being georegistered is converted to the coordinate system of the guiding component. If not checked, the coordinate system is not changed.

Interpolate pixels

Enabled when georegistering non-palette images or surfaces using the Numeric method. Creates a much smoother image when transforming the image into the new projection. Very computationally expensive: requires approximately twice the processing time to georegister an image if enabled. Note: Methods other than Numeric always interpolate pixels for non-palette images and surfaces and never interpolate pixels for palette images.

Scale pixels equally in X and Y direction

When this option is off, the user can specify both the width and the height of the resulting image or surface component. When the option is turned on, the user can only specify the width of the resulting component and the system will automatically compute the height. By default the option is turned on.

Save error surface using

Create a surface containing the root mean square error value for each location in the georegistered component. Choose the data type for the error number saved in the surface at each location. If checked, this option doubles the time required for georegistration. Enabled for the Numeric method. See the Error Surfaces topic.

Size

The transformed size of the image or surface in pixels. By default this will be set to some value that attempts to approximately preserve the size of the image or surface before georegistration. Enabled for images or surfaces.

 

The Order chosen will have a great impact on the number of control points required, especially when using the numeric method. For numeric georegistration, there usually must be at least four times the order number in control points, give or take a few. Thus, for numeric registration of order 4, there should be about 16 control points. This guideline is a minimum value. The Order value used can greatly effect the quality of the outcome. If the image is not georegistered well, try a different value of Order.

 

Certain arrangements of control points may require an even greater number of control points. Except for the Simple method, the more control points the better. Images of scanned paper maps will often be registered with 35 to 70 control points in a production map environment when using numeric or somewhat fewer points for affine registration.

 

Affine registration works with fewer control points, as few as two or three in the case of Affine (scale, shift, rotate). However, greater care must be taken when assigning affine control points. In general, control points should be placed for affine registration so that control points are drawn along the outer border of an imaginary shape without control points in the middle of the shape. For example, placing control points in a rectangular or rhomboidal arrangement is OK. Placing control points in a circle with several additional control points in the center of the circle is not OK.

 

Affine registration is a good choice when registering scanned images of maps that have a graticule grid of latitude and longitude lines shown. One can choose four intersections of latitude and longitude graticule lines that are near the four corners of the image to use as control points. Because fewer control points are required affine registration is often a better choice if we can place control points in a rectangular or rhomboidal arrangement.

 

Numeric registration may be a better choice when registering images to control points that are scattered throughout the image. When using numeric registration, evenly distributed control points will yield better results. This will, however, usually require many more control points than affine registration.

 

Either affine or numeric registration is a good choice when registering scanned images of maps that have a graticule grid of latitude and longitude lines shown if we are able to place many control points. One can choose a number of intersections of latitude and longitude graticule lines to rapidly mark many control points.

 

In either case, if control points are restricted to only part of the image it is quite likely that other parts of the image will not be georegistered well. When used in maps they will appear to be out of alignment with overlying drawing objects. It is critically important to use control points that are dispersed throughout the entire image. With the numeric method, the more control points that are used, the better the ultimate georegistration.

 

Given the labor of marking many control points the appeal of Simple registration is obvious. If an image is already in a North up overhead view and need only be resized and moved to be registered to a given target drawing this is a good choice. If the image is in Orthographic (the default when importing from geographically-unaware formats like .jpg), re-project the drawing into Orthographic and note the central latitude and longitude. Use Edit - Assign Projection with the image to specify the same central latitude and longitude. Then use Simple registration to georegister the image to the drawing.

 

Example

 

We begin with our SanFran sample Landsat image, which has been cropped to a smaller image centered upon San Francisco Bay.

 

images\img_numeric_points_01.gif

The image has had six control points added that are reasonably dispersed near the edges of the image. This is a good pattern when the image is a near-overhead view and numeric registration is to be used. However, in "real life" we would use many more control points, probably over 30 scattered throughout the image. The image will be registered to control points in a drawing of hydrographic lines imported from a 1:100K-scale DLG downloaded from USGS.

 

images\img_numeric_points_02.gif

After numeric registration, we can show the image in a map with the drawing lines show in yellow. Even though a small number of control points has been used registration is still very good.

 

Using Maps as Reference Components

 

If an image is being georegistered for display within a map it makes sense to use the map itself as a reference component, or to use a component that has the same projection as the map as the reference component. This will assure that the projection assigned to the image when it is georegistered will be exactly the same as that used by the map. The image will thus display faster within the map.

 

Maps are often a good choice for a reference component because they can show several layers such as roads and hydrography. Quite often a feature that is visible in an image may be identified with greater precision in a map when several layers are available to ascertain location.

 

images\sc_map_georeg_01.gif

 

For example, suppose we have an image like the one above where a bridge is visible. The example shows our SanFran sample image with the San Mateo Bridge crossing the South Bay.

 

images\sc_map_georeg_02.gif

 

The bridge provides a useful feature for use with control points, but if we work only with a hydro layer we will not see where the ends of the bridge could be marked.

 

images\sc_map_georeg_03.gif

 

Viewing both a roads layer and a hydro layer at the same time makes it easy to mark the ends of the bridge with control points. Note that the roads layer alone would not be any more useful than the hydro layer alone.

 

images\sc_map_georeg_04.gif

 

The resultant georegistration allows a reasonable match. The hydro outlines in the lower right of the image follow what USGS considers to be the shoreline in a region of tidal flats where a one foot (about a third of a meter) variation in water level changes the water's edge by hundreds or thousands of feet.

 

A Fast Arrangement

 

Many users will set up their consoles so that the reference drawing and the target image appear side-by-side with the control points pane in the middle.

 

images\sc_map_georeg_05.gif

 

One can then zoom into approximately the same region in both map and image windows to add control points.

 

images\sc_map_georeg_06.gif

 

Workflow with this arrangement is a simple cycle: zoom into particular region in both the map and the image. Click on the image and add three or four control points. Click on the map and add the analogous three or four control points. Click on the map and click the Back arrow to get back to the original view. Click on the image and click the Back arrow to get back to the original view. Repeat the cycle zooming into a different region to add control points.

 

Note that one can Zoom Box into a region, assign control points, and then click the Back arrow without losing the Zoom Box command. This sequence therefore is quick and easy to do. We can continue the zoom in / assign control points cycle until lots of control points have been assigned, and then georegister. With lots of control points the fastest method is usually Numeric.

 

Zooming into a particular region to place control points provides two main benefits:

 

·      It allows us to place several control points in that region with zoomed-in accuracy.

·      If the same region is zoomed in within the map window and the image window we are less likely to make a mistake when placing a control point because fewer points are involved.

 

The above advice, of course, assumes you have a reasonably high-resolution display with a reasonably large monitor so that there is enough room to have two reasonably useful windows open at once. 1024 x 768 is the absolute minimum, and really too small for serious georegistration work. Invest in a quality monitor or flat panel that's big enough to allow you to run at 1280 x 1024 resolution or greater.

 

Notes

 

Because control points can be imported from points that are in the drawing, a fast way of adding many control points is to add points into the drawing using Instant Data to specify the name of each point as it is clicked. We can then use these points as control points. See the Managing Control Points topic for essential methods that will save time and hassles.

 

Manifold will refuse to use control points that are placed on top of each other in the same image or drawing. The nature of control points is that they must refer to separate locations.

 

Most images that are overhead views of a subject can be georegistered to drawings in Orthographic projection using the Simple method with only two control points. This is a very fast match that is adequate for many GIS purposes.

 

The Save error surface as box option is not enabled for simple and affine methods.

 

Performance

 

Georegistering larger images can take substantial amounts of time. Depending on the size of image and the speed of the processor, large images can take hours to georegister. A very slow processor with an exceptionally large image might take days. The size of image, the speed of the processor and the method or other options selected can result in extensive variations in computation time required for georegistration. Checking the save error surface box will double the amount of time required.

 

For example, the Affine (triangulation) process is the most computationally intensive. Georegistering a 6500 x 7500 pixel grayscale image using the affine method on a slow machine with limited RAM, such as a 600Mhz PIII with 384MB of RAM, will take well over an hour. Georegistering the same image using the simple method (if it is a suitable overhead image that can be treated as Orthographic) to an Orthographic drawing will require a few seconds on the same machine. Georegistering the same image using the numeric method (less computationally intensive) on a faster machine (1.1 Ghz Athlon with 512MB of RAM) will require 10 to 15 minutes. Depending on the hardware and method chosen the time required to georegister the very same image can range from seconds to over an hour.

 

Try smaller images, or try georegistering the image using a greatly reduced number of pixels from the defaults initially appearing in the Register box. Begin with very small images until you have some experience in realistic time required for your system. Use a lower Order with the Numeric method to speed it up. See the Performance Tips topic for suggestions on getting maximum performance out of your system.

 

Consider also the true size of an image. An image that is imported from a compressed file format, such as a .tif file, that uses only 50MB on disk may actually represent over nearly a gigabyte of pixels. Compute the size of the image by multiplying the X and Y dimensions in pixels times four for RGB images and times five for RGBa images. Grayscale images require much less space.

 

Finally, check to see if the image has invisible pixels. It is almost always best to crop the image to the desired area of interest right after importing it into Manifold, to make sure that it is not bordered with a swath of invisible pixels. Doing so avoids wasting computational time on georegistering or otherwise manipulating large numbers of invisible pixels that have no real role in the image.

 

Technical Notes

 

The Numeric method computes a set of coefficients for X and Y polynomials of the specified order minimizing the error at control point locations. The computed polynomials are then applied to transform the source data.

 

The Affine (triangulation) method utilizes a proprietary manifold.net algorithm that deploys not one, but many affine transforms.

 

Large images or surfaces can take a very long time to georegister when using Affine (triangulation) method or the Numeric method with higher Order. If desired, the georegistration process can be cancelled and then re-launched using a different method or lower Order.

 

Occasionally we might make a mistake in the placement or naming of control points that results in a bizarre and obviously wrong georegistration. In such cases, it is nice to know that Undo will work.

 

Images must be resized during some types of georegistration. Palette images will always be resized using the nearest neighbor method (no interpolation of colors) to guarantee that the georegistration process does not introduce any new colors.

 

See Also

 

·      See the Manual Georegistration topic for discussion of georegistration by manual adjustment of projection properties.

·      See the Error Surfaces topic for examples of usage, including visual examples of the effect of higher Order and control point location.

·      For fine adjustment of registration, use the layer repositioning commands.

·      The File - Print dialog used to print a component includes a Control Points option that may be used to print the control points in a component.

·      The Georegister a Scanned Paper Map example topic shows a common georegistration task in step-by-step detail.