Applicability of the semi-automated photo-identification software ExtractCompare for the monitoring of the harbour seals in Ireland

Abstract

ABSTRACT: Photo-identification enables non-invasive identification of individuals from natural markings, with several algorithms developed to recognise individual patterns. ExtractCompare is a key software package for individual semi-automated recognition of harbour seals ( Phoca vitulina ). Its performance, however, may vary with image quality, aspect employed and subspecies. This study assessed ExtractCompare's applicability for monitoring harbour seals in Ireland where it has not been used previously, and explored the impact of aspect, pelage state and image parameters on performance. Test databases were created for the following aspects: head, neck, flank and head+neck. For each database, software performance was assessed by analysing 20 known matches and non-matches, respectively. The effect of parameters on performance was analysed using Generalised Linear Models, the Cumulative Density Function (CDF) was calculated for the first twenty ranked potential matching images, and similarity score thresholds were assessed against the relative frequency of false negatives. Results produced the first report of pelage state affecting the algorithm's accuracy, indicating that it should be considered for future studies. The CDF increased with the number of ranked images inspected (CDF1=0.35–0.60; CDF20=0.80–0.91). Our findings demonstrated the applicability of ExtractCompare for individual identification of harbour seals, indicating that this is a viable tool for monitoring this species in Irish waters.