Computer-assisted photo-identification of individual marine vertebrates: a multi-species system

A computer program named ‘ Finscan’ was developed for identifying individual marine animals by comparing new photographic images with a collection of previously identie ed images. The matching process was based on the pattern of nicks and notches commonly found along the trailing edge of the dorsal e n of many delphinid species. The program also allowed the inclusion of other user-dee ned descriptive features, such as leadingedge notches and truncated or irregular shapes. The output of the system was a presentation of images selected from the database, shown in order of similarity to a query image, so that the user could cone rm the match. Two algorithms for representing notch patterns were tested and compared and the system was evaluated with dorsal e n images of several marine vertebrates, as well as e uke images for one species. Using a database of images that were previously identie ed by expert observers, the performance of the system was measured in terms of the number of incorrect matches that were o Vered before the correct match. Since in most cases the correct match was oVered as the e rst or one of the e rst suggestions, the program substantially reduced the amount of eVort required to perform photo-based matching.