Automated digital individual identification system with an application to the northern leopard frog Lithobates pipiens

Identification of individual animals is needed for studying population demography, movement patterns and animal behavior. Animals with unique body markings (e.g., coloration or stripes) are easily identified by the human eye from photographs. However, automating this human ability with software is a complex task, compounded by various interdependent issues. This dissertation investigates critical issues of automating individual identification of northern leopard frogs Lithobates pipiens using images of their dorsal spot pattern. A standard image acquisition guideline was established to obtain high-quality photos. To identify individuals, a fingerprint area with the dorsal spot pattern is obtained from frog images. Image processing steps include fingerprint mapping, contrast normalization, spot extraction, and morphological operations. The spot pattern is represented with numerical features that form the basis for pattern comparison. Similarity measures were defined to compare spot patterns. A pattern recognition algorithm was developed to identify individual northern leopard frogs. The algorithm consists of multiple reduction steps that reduce false matching images at each step to return the top ten closest matches. An "Automated Animal Digital Identification System" (AADIS) was designed to register and identify individuals in the database. The system returns the top ten ranked individuals as closest matches along with a classification of these into probable versus not so probable categories. This classification serves as an indication whether the query individual is in the database or not. The system's performance was tested on different size datasets using random simulations. In a dataset of 200 individuals, with a total of 854 images, the system identified 95% of the true matches in the top ten and classified 87% of the true matches as probable in the top ten. Depending on the discriminators, the system classified 77-84% of individuals new to the database as not so probable matches. The developed pattern recognition algorithm may be adapted to other spotted animals by either adding or removing reduction steps suitable for the animal identification. AADIS is open-source software with a graphical user interface and a database. The image pre-processing steps take an average of 37 seconds for a trained user.