Abstract - In the food industry there are various foodstuffs in the
form of grains. Of particular importance is rice, being a
commodity crop. The ability to recognize defining characteristics
for identification is desirable as fraudulent mislabeling of rice
grain varieties is a growing problem In the present work a digital
imaging approach has been devised in order to investigate
different types of characteristics to identify different rice varieties.
Eight different common rice varieties were used in tests for
defining features. These include existing standards for grain
length and aspect ratio features, but also successfully show the
effectiveness of compactness as a feature. A novel texture feature
is also shown to be able to distinguish brown and milled rice in
greyscale images. All of these techniques are employed in an
inexpensive imaging system that is non-intrusive and nondestructive.
A highly effective yet simple imaging setup and
processing system is established, permitting image acquisition,
image processing, and feature extraction. Features are assessed
using unsupervised clustering techniques, showing the
dissimilarity between different varieties to a degree that would
allow successful identification.