Page 208 - Digital Analysis of Remotely Sensed Imagery
P. 208

172    Cha pte r  F i v e

               5.5.4  Creation of the Output Image
               The discussion so far has concentrated only on the geometric position
               of pixels in the rectified image. Their radiometric values in the output
               image have not been resolved yet. This issue is dealt with by resam-
               pling the input image. As illustrated in Fig. 5.9, pixels in the output
               image do not have a regular interval after the removal of geometric
               distortions. This pixel spacing does not reflect the spatial resolution
               of the original image anymore. Furthermore, the ground coordinates
               calculated using Eqs. (5.13) and (5.14) rarely correspond to the ground
               position where pixels are sampled. The output pixels do not have any
               radiometric values associated with them. Therefore, the idea of trans-
               ferring the value of pixels in the input image to the output image
               directly, as discussed previously, has to be abandoned. Instead, the
               process must be reversed for all pixels with the exception of the four
               corner ones. After their location in the output image is determined
               using the transformation equations. (5.13) and (5.14), the output
               image is created in two steps:

                    •  First, the position of all other pixels in the output image is
                      determined by sampling the newly created empty image at a
                      constant interval equivalent to the image’s spatial resolution
                      both horizontally and vertically, starting from the corner
                      pixels. In this way, it is guaranteed that the rectified image
                      will retain the same spatial resolution as the input image.
                      Once the position of all pixels in the output image is
                      determined, then their corresponding position in the input
                      image is estimated by inversing Eq. (5.15), or

                                         −1
                                     r = f (E, N)                   (5.21)
                                         1
                                         −1
                                     c = f (E, N)                   (5.22)
                                         2
                    •  Second, the radiometry at the position calculated in step one
                      in the input image is estimated via a process known as radiometric
                      resampling in which the radiometric value is estimated from the
                      pixel values in the neighborhood. Resampling of radiometric
                      values at these positions is precipitated by the fact that the
                      inversed  r and  c from Eqs. (5.21) and (5.22) are likely to be
                      floating points.  At these calculated positions there are no
                      pixels. Instead, radiometric values are available at pixels that
                      are spaced neatly in a regular grid in the original image. The
                      radiometric values at these calculated  r and  c coordinates
                      have to be estimated from the value of pixels in the vicinity.
                      The definition of the neighborhood size varies with the
                      resampling method.

                   There are three resampling methods in creating the output
               image.
   203   204   205   206   207   208   209   210   211   212   213