Classification and Retrieval of Images from Databases Using Rough Set Theory - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 193

while recall measures the number of relevant images retrieved, relative to the total number of relevant
images in the database. Table 1 displays the results for retrieval rates measured in terms of recall (Re),
precision (Pr), and F-measure (Fm) for the above-mentioned distance measures. We have partitioned
the databases into two classes: structure and non-structure. These two partitioned classes based upon
the measure of structure present in an image.
Figure 3 shows the comparative analysis between the three different similarity measures used in this
chapter plus the proposed one. We observe that the rough-based similarity measure is better in terms
of the number of retrieved and correct number of images in class.
It has been shown that a rough distance measure can lead to perceptually more desirable results
than Euclidean distance and histogram intersection methods as a rough distance measure considers the
cross similarity between features.
Figure 4 and 5 illustrate the overall classification accuracy in terms of total, retrieved, and correct
number of images in structure class compared with Decision tree, Discriminant analysis, Rough-neural
and rough set. Empirical results reveal that the proposed rough approach performs better than the other
classifiers. The number of generated rules before pruning was very large, which make the classification
slow. Therefore, it is necessary to prune the generated rules as we have done, and then the number of
rules has been reduced. Moreover, in the neural networks classifier, more robust features are required
to improve the performance of the neural networks classifier.
Figure 4. The overall classification accuracy