Abstract Many methods of classifying satellite imagery have been suggested. Most of the accepted methods, either supervised or unsupervised, are founded on the assumption that the radiance values of a particular ground cover type can be characterized by some probability distribution having a central value(s) and a range parameter(s). This assumption is rarely, if ever, tenable and hence the methods must produce unnecessarily high rates of misclassification. Thus there is a need for an approach which avoids this type of assumption. Pattern recognition techniques, e.g. the ‘watershed’ algorithm, can be used to divide up the radiance-frequency domain, provided the radiance values can be reduced to 2 dimensions. These methods do not presuppose any ‘shape’ for the radiance-frequency distributions and therefore provide an unbiased aid to the interpretation of satellite imagery.
[1]
Julius T. Tou,et al.
Pattern Recognition Principles
,
1974
.
[2]
T. M. Lillesand,et al.
Remote Sensing and Image Interpretation
,
1980
.
[3]
Jean Serra,et al.
Image Analysis and Mathematical Morphology
,
1983
.
[4]
Marcel J. E. Golay,et al.
Hexagonal Parallel Pattern Transformations
,
1969,
IEEE Transactions on Computers.
[5]
Serge Beucher,et al.
Use of watersheds in contour detection
,
1979
.
[6]
Philip N. Slater,et al.
Remote sensing, optics and optical systems
,
1980
.