International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
177 
(c) Inverse difference moment (d) Angular second moment 
Fig. 1. The four textural indices of a portion of the study area (the same area as that in Fig. 3 (a)). They were extracted from matrix 2 
with a 5x5 textural window. 
Then, the classification result with eighteen ordinal values was 
used to calculate the grey level co-occurrence matrix (matrix 2) 
for textural measure extraction. 
Four textural measures, i.e. contrast, entropy, angular second 
moment and inverse difference moment, were calculated from 
the two grey-level co-occurrence matrices (matrix 1 and matrix 
2). Four window sizes of 3 x 3 to 9 x 9 pixels were tested. The 
interpixel distance in the co-occurrence matrix calculation was 
one and the average of the four main interpixel angles (0°, 45°, 
90° and 135°) was used for the computations. 
2.3. Artificial Neural Networks Used in Classification 
A multi-layer perceptron (Benediktsson et al., 1990) was used 
as the ANN model (Fig. 2) in this study and a variant of the 
back-propagation algorithm (Paola and Schowengerdt, 1997), 
which is an enhanced version with a momentum term, was 
utilised to train the ANN model.