This paper presents an automated classification method
for images that represent UML Class diagram. To this end,
we discuss features extracted from images as input to the
classifier for UML class diagrams. In this study, we introduced
23 features that capture statistical and geometric characteristics
of diagrams. We find that using these metrics as
predictors for the classification reaches 95.9% and 91.4%
(respectively) of correct classification of input images for
UML CD and non-UML CD. For this study 1300 different
images are collected from the Internet through Google Image
Search. We make this dataset available as a benchmark.