APHIS: A new software for photo-matching in ecological studies
Introduction
Detailed data on individual life-history are used in ecological and evolutionary studies for the estimate of demographic parameters such as population size, survival and fertility of wildlife populations (e.g. Fernández-Chacón et al., 2011, Lebreton and North, 1993, Tavecchia et al., 2001, Tavecchia et al., 2005, Williams et al., 2001). A common solution for the individual recognition of the animals is to apply a mark to the animal body in the form of a tag or a ring with a unique alphanumeric code. However, rings, tags, flipper bands or other marks can alter individual fates and behavior (Gauthier-Clerc et al., 2004, McCarthy and Parris, 2004). In addition to ethical issues (e.g. May, 2004), these negative effects lead to bias the estimates of the parameters of interest. As a consequence there is an increasing interest in using non-invasive methods for individual recognition, such as unique natural marks or body characteristics. These methods have been applied with success in a wide range of taxa, in mammals (Karanth and Nichols, 1998, Langtimm et al., 2004, Martínez-Jauregui et al., 2012), amphibians (Gamble et al., 2008), reptiles (Sacchi et al., 2010), fishes (Speed et al., 2007, Van Tienhoven et al., 2007) or cephalopods (Huffard et al., 2008). However, with few exceptions (i.e. Perera et al., 2001), the photo-identification is restricted to those species featuring distinct colors, spots or marks. Photo-identification procedures consist of comparing a sample picture of an unknown individual with a library of candidate images of previously photographed individuals. This search is, in many cases, conducted by experienced observers who compare patterns and scars between photographs with the naked eye and might be extremely time-consuming when library contains hundreds of images (e.g. Martínez-Jauregui et al., 2012, Verborgh et al., 2009). Naked-eye comparisons are typically assisted by a preliminary grouping of the images using a multi-character score, for example by grouping images with a given chromatic pattern (e.g. absence or presence of specific marks, Carafa and Biondi, 2004). Unaided procedures may also become prone to errors when image libraries expand. There is now a growing demand in developing automatic or computer-aided procedures for photo-matching (Gamble et al., 2008). A computer-aided photo-identification system identifies the most probable sample–candidate matches, reducing the number of images to be inspected. Most photo-identification software solutions concatenate three processing steps. The first is a preprocessing step where a region of interest is selected and the image rotated, scaled or spatially corrected if required by comparison algorithms; the second is usually an automated comparison between the sample and the library of images, which arranges candidates by matching probability or likelihood values; a final step is a visual comparison of sample–candidate pairs for a limited number of plausible matches.
We present a new software solution, APHIS (Automated PHoto-Identification Suite), specially designed to deal with sample sets of over a hundred photographs per field campaign and image libraries containing more than a thousand samples. APHIS proposes two approaches for photo-matching, the Spot Pattern Matching (SPM) and the Image Template Matching (ITM). The former has been built on the already existing I3S algorithm (Van Tienhoven et al., 2007) while the latter is a novel approach based on pixel matching that minimizes the user's preprocessing effort. ITM is a fast-running alternative to study species with apparent or easily recognizable spots or colored parts of the skin. The workflow and graphic interface of APHIS have been designed to reduce the time invested by the researcher in analytical tasks and to enhance user experience. We describe below the general features of the APHIS interface and illustrate the SPM and ITM procedures using real data from two capture-photo–recapture studies on the Balearic Lizard, Podarcis lilfordi, and on the Northern spectacled salamander, Salamandrina perspicillata (Fig. 1).
Section snippets
Automated PHoto-Identification Suite (APHIS)
APHIS (Automated PHoto-Identification Suite, freely available at http://www.imedea.uib-csic.es/bc/ecopob/) v. 1.0 combines C++ and Java modules. The idea behind APHIS was to provide users with a flexible environment for photo handling and matching. The Graphic User Interface (GUI) has been programmed using the Nokia Qt framework (http://qt.nokia.com/). The image preprocessing and analysis of the ITM approach implements functions from the openCV v. 2.2 libraries (Bradski, 2000). The two
Results
The 287 images, of which 91 were recaptures, were analyzed using the SPM and the ITM procedures. The two approaches, ITM and SPM, delivered similar results, however, the overall number of photos classified as new captures by both approaches were different. The SPM approach correctly classified all newly photographed individuals (percentage of correctly classified pictures = 100%), while ITM found 85 of the 91 recaptures (93.4%). Excluding user's mistakes (e.g. reference points placed wrongly),
Discussion
Individual identification by photo recognition is becoming an increasing area of research. At present, there are several routines available for photo-matching, for example ‘I3S’ (van Tienhoven et al., 2007),‘MantaMatcher’ (Town et al., 2013), ‘StripeSpotter’ (Lahiri et al., 2011), Sloop (Gamble et al., 2008) and ‘Wild-ID’ (Bolger et al., 2012). Some are highly customized and some are very flexible. Our purpose here was neither to compare them nor to create yet another procedure for a particular
Acknowledgments
We thank all the people that have helped in collecting animal images. In particular, Marco Basile, Andrea Costa, Aldo Crisci, Daniele Scinti-Roger, Mario Posillico, Rodolfo Bucci, Filippo Della Civita and all members of the Population Ecology Group at the IMEDEA. The Northern spectacled salamander project was supported by the MANFOR CBD project (LIFE09 ENV/IT/000078). We are grateful to L. Bonnet for her help with logistics at Moltona island.
Capture permits of S. perspicillata were obtained
References (31)
- et al.
Population abundance, size structure and sex-ratio in a insular lizard
Ecol. Model.
(2013) - et al.
A computer-assisted system for photographic mark–recapture analysis
Methods Ecol. Evol.
(2012) The OpenCV library
Dr. Dobbs J.
(2000)An Analysis of Utilizing the Leatherback's Pineal Spot for Photo-identification
(2008)- et al.
Application of a method for individual photographic identification during a study on Salamandra salamandra gigliolii in central Italy
Ital. J. Zool.
(2004) - et al.
Spatial heterogeneity in the effects of climate change on the population dynamics of a Mediterranean tortoise
Glob. Chang. Biol.
(2011) - et al.
Multi-scale features for identifying individuals in large biological databases: an application of pattern recognition technology to the marbled salamander Ambystoma opacum
J. Appl. Ecol.
(2008) - et al.
Long-term effects of flipper bands on penguins
Proc. R. Soc. B Biol. Sci.
(2004) - et al.
Individually unique body color patterns in octopus (Wunderpus photogenicus) allow for photoidentification
PLoS One
(2008) - et al.
Texture classification with generalized Fourier descriptors in dimensionality reduction context: an overview exploration
Estimation of tiger densities in India using photographic captures and recaptures
Ecology
Biometric animal databases from field photographs: identification of individual zebra in the wild
Survival estimates for Florida mantees from the photo-identification of individuals
Mar. Mammal. Sci.
Distinctive image features from scaleinvariant keypoints
Int. J. Comput. Vis.
Cited by (58)
Photo identification for sea turtles: Flipper scales more accurate than head scales using APHIS
2023, Journal of Experimental Marine Biology and EcologyHuman expertise combined with artificial intelligence improves performance of snow leopard camera trap studies
2023, Global Ecology and ConservationCitation Excerpt :In recent years, AI and deep learning methods have demonstrated significant improvements in identifying individual animals from photographs and capture-recapture studies. Researchers have evaluated software programs with varying algorithmic performance, such as Wild-ID (Bolger et al., 2011), I3S Pattern+ (Hartog and Reijns, 2013), APHIS (Moya et al., 2015), HotSpotter, SIFT-based algorithms, and AmphIdent (Matthé, 2015), to identify amphibians, harbor seals (Phoca vitulina vitulina), Masai giraffe (Giraffa camelopardalis tippelskirchi), and wildebeest (Connochaetes taurinus) (Bardier et al., 2020; Matthé et al., 2017; Dawson, 2021). Different systems have yielded varying degrees of precision for various taxa.
Fingerprinting pike: The use of image recognition to identify individual pikes
2020, Fisheries ResearchCitation Excerpt :The image matching can be done by the naked eye, a process that is subjective and time-consuming (Matthé et al., 2017), or with the use of software. Several software solutions have been developed, and the three most widely used are freely available: APHIS (Moya et al., 2015), Wild-ID (Bolger et al., 2012), and I3S Pattern+ (Van Tienhoven et al., 2007). While APHIS and Wild-ID are both fully automatic, meaning that they automatically extract and interpret pattern features, the user defines the patterns to be used by I3S Pattern+ (Sacchi et al., 2016).
Evaluating the performance of semiautomated photographic identification programs for leopard seals
2024, Wildlife Society Bulletin