of tonal and textural. variables at the three scales were further
analysed. Linear discriminant analysis was again carried out on
the data by combining the two sets of corn (cut corn and uncut
corn) as one group. Pairwise linear discriminant scores were also
obtained for the more significant variables (as determined earlier)
and their suitable combinations.
Summary of Results and Discussions
The above analysis enabled testing of the effectiveness of each
variable and their combinations to further evaluate the relative
contribution of each variable marginally under different combi-
nation of crop types. Utility of texture as a supplement to tone
under these varied conditions could thus be evaluated.
The discriminant scores as a result of the discriminant analyses
carried out (for all 5 crop types) are summarised in Table 3.
Table 3
Summary of Percentage of Successful Scores Using Tone/Texture
at Three Scales
Scale Tone Texture Tone+Tex. Tone+Tex. Tone+Texture
only only (Linear) (Quad. ) (All three scales)
(Linear) (Quad.)
Small 42$ 43$ 2% $
Medium 45$ 45$ $ 2 80$ 66$
Large 49$ % 69% $
The table clearly illustrates that by combining tone and texture
at any of the three scales, success of discrimination improves
very significantly. It also shows that by using all the three
scales, still better success is obtained. An optimum scale for
best discrimination is noticeable. As these optimums for tone
and texture are not the same, the improvement in success by use
of more than one scale is further explained.
Summary of results for various combinations of the six variables
is given in Table 4 for both cases i.e. with all five crops and
with four crop types (the two corn states merged into one crop
type).
Table 4 illustrates that:
a) there is a marked improvement in overall success by aggregating
the corn category into a single category;
b) use of texture in addition to tone and use of more than one
scale of photographs should contribute to significantly better
success of classification;
c) the combined use of the four variables of tone and texture at
medium and large scale provide the best results.
Lastly , the role of texture and tone for pairwise discrimination
is summarised in Table 5. This supportive role of texture in the