Semi-Automatic Identification of Humpback Whales

This paper describes current work on a photo-id system for humpback whales. Individuals of this species can be uniquely identied by the light and dark pigmentation patches on their tails. We propose semi-automatic algorithm based on marker-controlled watershed transformation for segmenting the animal's tail from the surrounding sea. We propose tting an afne invariant coordinate grid to the resulting segmentation. The grid can be adjusted according to the level of occlusion by the sea. A numerical feature vector capturing the patch-distribution with respect to the grid is then automatically extracted and used to match the individual against the database of similarly processed images.