Animal recognition application using Kohonen Feature Map / Muhamad Syafiq Nordin

Underwater world had always been full of mystery in view of the fact that it was filled
with unaccountably many species. Among the living organisms, fish are the most familiar
to humans in environment, commercial and even recreational. From this perspective, fish
recognition arouses interest of not only dedicated underwater scientists but also of
ordinary people who may be interested in this matter. Roughly 1.4 million species are
known to science. Beyond this estimation, most unrecognized species are in poorly
studied groups where it habitats were seldom explored. The job of discovering new
species falls on the area of biology called taxonomy. The World-Wide Web is being used
to collect data used by taxonomists' for instance taxonomic literature and specimen
databases in different parts of the globe, archived as digital images. This scenario had
shown us that there is a need for an animal recognition tool that supports efficient
searching and navigating through large image databases of specimens. In this research, a
prototype of animal recognition application using Kohonen Feature Map was introduced.
The system has a learning component that able to classify fish species based on the local
visual feature of its representative image. This research also reveals Kohonen Feature
Map as a promising tool for image classification. Realized that there is millions of
species around the globe, this research focused on fish species that was common in
Malaysia. 20 species were studied in this research. The image database used in the
research was composed of 100 color images