Page 278 - Digital Analysis of Remotely Sensed Imagery
P. 278

240    Cha pte r  S i x


           Band  1         2         3        4         5        7
           1       21.7564
           2        8.6635  5.1738
           3       11.3633  5.8163   12.1910
           4      −11.7377  0.0769    9.8601  132.3204
           5       −1.0364  3.1006   14.5452   83.5923  74.5963
           7        3.6701  2.8159    8.6601   26.9379  28.8602  15.0311
          TABLE 6.3  Variance-Covariance Matrix of the Auckland TM Subscene Image in Fig. 6.23a


               band 4, and the least informative is band 2. Its small variance means
               that most pixels have a similar value in this band, making it difficult
               to interpret the image.
                   Similar to the variance-covariance matrix, the correlation matrix
               (Table 6.4) is also square and symmetric. Again, its dimension is the same
               as the number of spectral bands used in the analysis. Unlike the
               variance-covariance matrix, all main diagonal values have a value of 1
               as the content of each band is perfectly correlated with itself. All the
               off-main diagonal values vary between −1.0 and 1.0. Virtually, these
               correlation coefficients represent the degree of data redundancy between
               any two bands. For instance, the information content of bands 4 and 5 is
               highly correlated with each other at 84.14 percent. This means more than
               three quarters of the information is shared between the two bands. This
               high correlation, however, does not imply that band 4 is of little value as
               its degree of correlation with band 2 is very low at only 0.30 percent.
                   During PCA it is possible to output only a portion of all possible com-
               ponents (Fig. 6.23). The first three component images (Fig. 6.23b, c, and d)
               are still informative in a sense that the scene is illustrated quite well.
               However, starting from the fourth component (Fig. 6.23e), the quality of
               the component images deteriorates so drastically that little information
               about the scene is preserved. It is not absolutely necessary to output all the


            Band   1         2         3         4        5        7
            1       1
            2       0.8166   1
            3       0.6977   0.7323    1
            4      −0.2188   0.0030    0.2455    1
            5      −0.0257   0.1578    0.4823    0.8414   1
            7       0.2029   0.3193    0.6397    0.6040   0.8619   1

          TABLE 6.4  Correlation Matrix of the Auckland TM Subscene Image in Fig. 6.23a
   273   274   275   276   277   278   279   280   281   282   283