Parent Topic: GCPWORKS

Disk To Disk Registration

The ``Disk to Disk Registration'' panel is used to perform final registration from the uncorrected image file to a final output (geocoded) image file. Any number of channels may be registered at once, to any desired output file with the appropriate georeferencing type. The registration option may be modified first. This panel is launched by selecting ``Perform Registration to Disk'' from the main GCPWorks control panel.

The Disk to Disk Registration panel consists of four subareas, the first three of which deal with channel/file mapping selection and the fourth of which is used to set registration options.

Once the desired set of registration options has been set, and the desired input/output channel mapping established, the ``Perform Registration'' button at the bottom of the panel is used to initiate the registration. Disk to disk registration can take a lot of time for input and/or output files. If the files used are very large, and the machine running GCPWorks has a lot of memory, increase the size of the main memory caching arrays using the ``Memory'' control field on the panel.

For each pixel in the output image, the appropriate pixel(s) is located and read in from the input image. Thrashing occurs when large amounts of time are spent reading in data to find each individual input pixel. The only way to avoid thrashing is to increase the main memory caching array; but, if this is increased to much larger than the physical memory available for the machine on which GCPWorks is run, performance of the OS may degrade seriously, and GCPWorks would not be significantly helped, if at all.

While the registration is running, a progress bar will be displayed. If the registration algorithm determines that the registration is ``thrashing'', it will start the registration over again, operating on strips of the output image at a time, and the progress bar will restart at 0% done.

Channel/File Mapping Selection

The first three sub-areas of the Disk To Disk Registration panel are used to select an output file, and to select input and output channels for the registration.

If an output file does not already exist, select ``New Output File'' and type the name of the new file to create. A series of dialogs will appear, allowing the creation of an output PCIDSK file.

During Image To Image Registration, the output file will default to the georeferenced database used for tiedown.

To select another image file into which to register/mosaic, select the ``New Output File'' button. The current output file is shown in the ``Selected Channel/File Mappings'' area after the ``Output File:'' prompt.

A list of the channels available for selection on the uncorrected image file is available under the title ``Uncorrected Image Channels''. The list of channels available for selection on the output file is shown under the title ``Output File Image Channels''. The current input to output channel mapping is shown in the ``Selected Channel/File Mappings'' area after the prompt ``Channel Mapping''.

To select a new channel mapping click on the ``Reselect'' button and then select each channel input/output pair from the ``Uncorrected Image Channels'' and ``Output File Image Channels'' lists.

Resampling Mode

The ``Resampling Mode'' determines the type of resampling done during the registration process. Possible choices are nearest neighbour, bilinear interpolation, (approximate) cubic convolution, 8pt sin(x)/x or 16pt sin(x)/x.

Model Order

The ``Model Order'' determines the order of the polynomial used to perform the registration warping. For instance, if the model order were three, a cubic polynomial would be used to model the warping.

Each model order above first has a required minimum number of GCPs. The minimum numbers of GCPs for each order from two to five are summarized below:

         Required GCPs                Order
               7                       2nd
              11                       3rd
              16                       4th
              22                       5th

Background

The ``Background'' editable field is used to set the background value for mosaicking purposes. This is a data value which indicates invalid or missing data. Pixels of uncorrected image data with this value will not be registered into the georeferenced image. Georeferenced image pixels with this value will always be completely replaced by the registered uncorrected image value if they are within blend width of the mosaic cut line.

If this value is not set, there is no background value. Many satellite images are delivered with areas of missing data at the edges. These areas typically have a value of `0', so a background value of 0 would avoid treating these areas as real data.

Colour Matching

The ``Colour Matching'' toggle switch is used to enable or disable use of the radiometric correction LUT(s) created in the Colour Matching step. If the Colour Matching step were not performed, the toggle switch would be insensitive.

Colour Matching is currently only available when registering the uncorrected image channels which were used during the tie down and colour balancing process.

Blend Width

The ``Blend Width'' field slider is used to control the distance around the mosaic cut line for which blending of the georeferenced and uncorrected image is performed.

Blending performs a form of feathering around the cut line, gradually changing from 100% georeferenced image data values to 100% uncorrected image data values. The blend distance determines the distance to either side of cut line over which blending is performed. By default it is 0, indicating that no blending is performed.

Memory Cache

Near the bottom of the panel is a slider and text field that can be used to control the size of the cache array used when performing the registration. The default size of the cache is 4MB (about 4 million bytes), but increasing the cache size dramatically improve performance on workstations with more than 16MB of physical memory.

The value entered in the text field is measured in MB (mega-bytes) and a reasonable value would one half the physical memory available on the computer being used.