Page 205 - Digital Analysis of Remotely Sensed Imagery
P. 205

170    Cha pte r  F i v e

                    •  Second, the order of transformation is raised if there is a
                      sufficient number of GCPs available. Caution must be
                      exercised in adopting a higher order in that this order of
                      transformation must have a physical meaning. Furthermore,
                      the increase in the transformation order has its limits.
                      Beyond the second order the accuracy of transformation
                      improves only marginally despite an exponential increase
                      in the amount of work involved. Besides, higher-order
                      polynomials produce unreliable results for satellite images
                      with simple geometric conditions (e.g., near-vertical or
                      relatively flat areas), while a low-order polynomial is
                      able to produce submeter rectifications (Rosenholm and
                      Akerman, 1998).

                   Thus, the second option is not as effective as the first one. In either
               case, the transformation coefficients have to be recalculated with the
               revised georeferencing setting, and the RMSE updated before they
               are used to create the output image.
                   Shown in Table 5.3 is an example of image rectification results
               using 17 GCPs. The first column represents GCP sequential numbers.
               GCP image coordinates expressed in row and column are provided
               in the second and third columns, respectively. They can be entered
               into the computer by clicking on the points in the image directly.
               Their coordinates in the ground coordinate system to be projected
               are listed in the next two columns. They have to be entered into the
               computer manually. The nature (type) of GCP is identified in the
               next column. It can have two possibilities, control or check. In this
               particular case, all GCPs were used as control points. Calculated
               using Eqs. (5.18) and (5.19), respectively, the residuals in easting and
               northing for each GCP are provided in columns seven and eight,
               expressed in the image unit (e.g., pixel size). The last column shows
               the RMSE at each point, calculated using Eq. (5.20) with n being 1.
               Those with the largest residual or largest contribution represent the
               worst GCPs. For instance, the unusually large residual at GCP 15 in
               easting (4.699) is indicative of the presence of a mistake that could
               have stemmed from incorrect entry of the coordinates into the com-
               puter or erroneous identification of the GCP on the image or on the
               ground. This GCP may be discarded by turning it into a check point.
               Of particular note is the connectivity among all residuals. After the
               largest residual associated with GCP 15 is removed from the calcu-
               lation, residuals at other GCPs will become smaller accordingly. Y
               residuals will become larger for some GCPs because its Y residual
               (0.157) is smaller than the mean. Also presented in the table are the
               overall residuals in both easting (RMSE   = 1.4957) and northing
                                                   X
               (RMSE = 0.4936), with the overall RMSE  being 1.575. This out-
                     Y                              XY
               come, larger than the expected 1, can be reduced to within one pixel
               after GCP 15 is turned into a check point.
   200   201   202   203   204   205   206   207   208   209   210