View -invariant curve and point -pattern matching with application to photo-identification of marine mammals

Individual identification of marine mammals is an important problem in the field of marine mammal biology. The recognition of individual animals provides vital information about their social and ecological patterns. Conventional methods for identification, like tagging and heat-branding, suffer from practical and legal obstacles. Photo-identification of marine mammals utilizing their natural markings has been widely used in field biology. Large photographic databases have already been created by marine mammalogists. However, most laboratories across the world still perform the identification task by manual inspection of the photographs, wherein a query image is compared to each image from a database of photographs until a suitable match is found. As the number of identified individuals within a study population increases, this process becomes labor-intensive and increasingly subject to human-error. The goal in this research effort has been the development of computer-assisted semi-automated techniques for individual identification of marine mammals via pattern recognition methods.;Natural and unique markings found on the animals' bodies can be used as identifying features. Two such features that have been utilized in this dissertation are curves and patches found on the animals' bodies. Image processing techniques are used for image pre-processing and extraction of these features. Pattern matching techniques are then applied to these features in order to retrieve the closest matches from a previously cataloged database to a given query image. This dissertation discusses novel curve and point-pattern matching techniques to accomplish this objective.;Image databases of 4 different marine mammals, namely dolphins, sea-lions, gray whales and humpback whales, have been used to examine and compare the performance of the existing photo-identification techniques with those previously developed. The performance results indicate that the developed methods offer a significant advantage over the previous ones in terms of recognition rate. The outcome of this work is applicable in several branches of marine mammal biology. Finally, it should be noted that although the main focus of this dissertation has been marine mammal identification, the pattern matching techniques developed here have a broader applicability, that is, they can be used for other pattern recognition applications.