Warp

Warp Operator


The Warp operator is a component of coregistration. This operator computes a warp function from the master-slave ground control point (GCP) pairs produced by GCP Selection operator, and generates the final co-registered image.

Compute Warp Polynomial

   Once the valid master-slave GCP pairs are known, a polynomial of a certain order is computed using a least square method, which maps the master GCPs onto the slave GCPs. This function is known as the warp function and is used to perform the co-registration. Generally the warp function computed with the initial master-slave GCPs is not the final warp function used in co-registration because it normally introduces large errors for GCPs. These GCPs must be removed. Therefore the warp function is determined in an iterative manner.
  1. First a warp function is computed using the initial master-slave GCP pairs.
  2. Then the master GCPs are mapped to the slave image with the warp function, and the residuals between the mapped master GCPs and their corresponding slave GCPs are computed. The root mean square (RMS) and the standard deviation for the residuals are also computed.
  3. Next, the master-slave GCP pairs are filtered with the mean RMS. GCP pairs with RMS greater than the mean RMS are eliminated.
  4. The same procedure (step 1 to 3) is repeated up to 2 times if needed and each time the remaining master-slave GCP pairs from previous elimination are used.
  5. Finally the master-slave GCP pairs are filtered with the user selected RMS threshold and the final warp function is computed with the remaining master-slave GCP pairs.
  
    The WARP polynomial order is specified by user in the dialog box.

Generate Co-Registered Image

   With the determination of warp function which maps pixels in the master image to the slave image, the co-registered image can be obtained with interpolation. Currently the following interpolation methods are supported:
  1. Nearest-neighbour interpolation
  2. Bilinear interpolation
  3. Bicubic interpolation
  4. Cubic interpolation (4 and 6 points)
  5. Truncated sinc interpolation (6, 8 and 16 points)

Interpolation for InSAR

    For interferometric applications Cubic or Truncated sinc kernels are recommended. These kernels assure the optimal interpolation in terms of Signal-to-Noise ratio.

Residual File

   The residual file is a text file containing information about master and slave GCPs before and after each elimination. The residual for a GCP pair is the errors introduced by the warping function and can be used as a good indicator of the quality of the warp function. It is often very useful to check the information contained within the residual file to see if the co-registration process can be considered to have been successful. For example, the "RMS mean" value can be used as an approximate figure of merit for the co-registration. User can view the residual file by checkmarking the "Show Residuals" box in the dialog box. Detailed information contained in the residual file are listed below:

Parameters Used

   The following parameters are used by the operator:
  1. RMS Threshold: The criterion for eliminating invalid GCPs. In general, the smaller the threshold, the better the GCP quality, but lower the number of GCPs.
  2. Warp Polynomial Order: The degree of the warp polynomial. 
  3. Interpolation Method: The interpolation method used computing co-registered slave image pixel value.
  4. Show Residuals: Display GCP residual file if selected.