results in 32x8 dimensional feature vector. The average cell area Aµ was calculated for
every image. All the features were normalised to have zero mean and unit variance.
Figure 4 shows the total misclassification percentage together with SE, SP, PvP, and
PvN measurements plotted against increasing K in Knn-d classifier. The results are a
validation of the overall process and performance comparison of the different features.
An examination of the total classification error (Figure 4(a)) indicates the most successful
feature is C + M + R followed in the order by H + M + R, C, and H. Thus, feature C is
more successful than H. However, the difference is not significant especially considering
the cost in calculation time for C. The performances of M and R are low which suggests
that colour information is essential. However, they provide slight boosts for the H and
C features. To interpret the results in more detail, if we choose the value K = 3 in the
feature (C+M +R), the SE ( 74%) performance value reveals the probability of the result
being positive given that the stained object is a parasite. The SP ( 98%) value reveals
that the probability of the result being negative given that the object is not a parasite.