228
9: Chintan A. Shah
d
Fig.9.3a-d. First six features extracted by a PCA, b SPCA, c OSP, and d PP applied on the
AVIRIS data. The features extracted by PCA and SPCA, are ranked in decreasing order of
variance. For OSP and PP, the features are ranked in the order of projections
obtained through SPCA capture significant class information. A visual inspection of the features generated by OSP reveals that its performance is comparable
to that of PCA and SPCA in its ability to discriminate between classes. Of particular interest is the third feature generated by OSP (Fig. 9.3c), in which it
detects the signature corresponding to class background. For PCA and SPCA,
9: Chintan A. Shah
d
Fig.9.3a-d. First six features extracted by a PCA, b SPCA, c OSP, and d PP applied on the
AVIRIS data. The features extracted by PCA and SPCA, are ranked in decreasing order of
variance. For OSP and PP, the features are ranked in the order of projections
obtained through SPCA capture significant class information. A visual inspection of the features generated by OSP reveals that its performance is comparable
to that of PCA and SPCA in its ability to discriminate between classes. Of particular interest is the third feature generated by OSP (Fig. 9.3c), in which it
detects the signature corresponding to class background. For PCA and SPCA,