GCPs are specific pixels in an image for which the output map coordinates (or other output coordinates) are known. GCPs consist of two X,Y pairs of coordinates:
- source coordinates—usually data file coordinates in the image being rectified
- reference coordinates—the coordinates of the map or reference image to which the source image is being registered
The term map coordinates is sometimes used loosely to apply to reference coordinates and rectified coordinates. These coordinates are not limited to map coordinates. For example, in image-to-image registration, map coordinates are not necessary.
GCPs in ERDAS IMAGINE
Any ERDAS IMAGINE image can have one GCP set associated with it. The GCP set is stored in the image file along with the raster layers. If a GCP set exists for the top layer that is displayed in the Viewer, then those GCPs can be displayed when the Multipoint Geometric Correction tool (IMAGINE ribbon Workspace) or GCP Tool (Classic) is opened.
In the CellArray of GCP data that displays in the Multipoint Geometric Correction tool or GCP Tool, one column shows the point ID of each GCP. The point ID is a name given to GCPs in separate files that represent the same geographic location. Such GCPs are called corresponding GCPs.
A default point ID string is provided (such as GCP #1), but you can enter your own unique ID strings to set up corresponding GCPs as needed. Even though only one set of GCPs is associated with an image file, one GCP set can include GCPs for a number of rectifications by changing the point IDs for different groups of corresponding GCPs.
Accurate GCPs are essential for an accurate rectification. From the GCPs, the rectified coordinates for all other points in the image are extrapolated. Select many GCPs throughout the scene. The more dispersed the GCPs are, the more reliable the rectification is. GCPs for large-scale imagery might include the intersection of two roads, airport runways, utility corridors, towers, or buildings. For small-scale imagery, larger features such as urban areas or geologic features may be used. Landmarks that can vary (for example, the edges of lakes or other water bodies, vegetation, and so forth) should not be used.
The source and reference coordinates of the GCPs can be entered in the following ways:
- They may be known a priori, and entered at the keyboard.
- Use the mouse to select a pixel from an image in the Viewer. With both the source and reference Viewers open, enter source coordinates and reference coordinates for image-to-image registration. The Multipoint Geometric Correction tool contains the both the source and reference Viewers within the tool.
- Use an existing Ground Control Coordinates file (.gcc file extension). This file contains the X and Y coordinates along with the GCP point ID, saved as an external file.
- Use a digitizing tablet to register an image to a hardcopy map.
Digitizing Tablet Option
If GCPs are digitized from a hardcopy map and a digitizing tablet, accurate base maps must be collected. You should try to match the resolution of the imagery with the scale and projection of the source map. For example, 1:24,000 scale USGS quadrangles make good base maps for rectifying Landsat TM and SPOT imagery. Avoid using maps over 1:250,000, if possible. Coarser maps (that is, 1:250,000) are more suitable for imagery of lower resolution (that is, AVHRR) and finer base maps (that is, 1:24,000) are more suitable for imagery of finer resolution (that is, Landsat and SPOT).
When entering GCPs with the mouse, you should try to match coarser resolution imagery to finer resolution imagery (that is, Landsat TM to SPOT), and avoid stretching resolution spans greater than a cubic convolution radius (a 4 × 4 area). In other words, you should not try to match Landsat MSS to SPOT or Landsat TM to an aerial photograph.
How GCPs are Stored
GCPs entered with the mouse are stored in the image file, and those entered at the keyboard or digitized using a digitizing tablet are stored in a separate file with the extension .gcc. GCPs entered with the mouse can also be saved as a separate *.gcc file.
GCP Prediction and Matching
Automated GCP prediction enables you to pick a GCP in either coordinate system and automatically locate that point in the other coordinate system based on the current transformation parameters.
Automated GCP matching is a step beyond GCP prediction. For image-to-image rectification, a GCP selected in one image is precisely matched to its counterpart in the other image using the spectral characteristics of the data and the geometric transformation. GCP matching enables you to fine tune a rectification for highly accurate results.
Both of these methods require an existing transformation which consists of a set of coefficients used to convert the coordinates from one system to another.
GCP prediction is a useful technique to help determine if enough GCPs have been gathered. After selecting several GCPs, select a point in either the source or the destination image, then use GCP prediction to locate the corresponding GCP on the other image (map). This point is determined based on the current transformation derived from existing GCPs. Examine the automatically generated point and see how accurate it is. If it is within an acceptable range of accuracy, then there may be enough GCPs to perform an accurate rectification (depending upon how evenly dispersed the GCPs are). If the automatically generated point is not accurate, then more GCPs should be gathered before rectifying the image.
GCP prediction can also be used when applying an existing transformation to another image in a data set. This saves time in selecting another set of GCPs by hand. Once the GCPs are automatically selected, those that do not meet an acceptable level of error can be edited.
In GCP matching, you can select which layers from the source and destination images to use. Since the matching process is based on the reflectance values, select layers that have similar spectral wavelengths, such as two visible bands or two infrared bands. You can perform histogram matching to ensure that there is no offset between the images. You can also select the radius from the predicted GCP from which the matching operation searches for a spectrally similar pixels. The search window can be any odd size between 5 × 5 and 21 × 21.
Histogram matching is discussed in Enhancement.
A correlation threshold is used to accept or discard points. The correlation ranges from -1.000 to +1.000. The threshold is an absolute value threshold ranging from 0.000 to 1.000. A value of 0.000 indicates a bad match and a value of 1.000 indicates an exact match. Values above 0.8000 or 0.9000 are recommended. If a match cannot be made because the absolute value of the correlation is less than the threshold, you have the option to discard points.