seaturtle.org : MTN : ARCHIVES : Sign In

Methods used for computer aided sea turtle photo-ID include matching of head scales using numerically coded scale shape patterns (Jean et al. 2010; Valdés et al. 2014), Interactive Individual Identification System (I³S) Classic (Dunbar et al. 2014) and subjective pattern matching (Bennett & Bennett 2001). Stability of facial scale arrangement and shape has been shown to be stable for up to 11 years (Carpentier et al. 2016). Facial scale Photo- ID does have significant drawbacks for in-water acquired images. To obtain a useful image the turtle should be approached from the side and parallel to the turtle with less than a 20° viewing angle (Araujo et al. 2016) limiting the use of citizen scientist snorkelers. Tumors can also obscure the facial scales (<http://www.turtles.org/identify.htm>).

Few examples of investigating flipper scale patterns for Photo- ID exist (Caillouet et al. 1985; Gatto et al. 2018; <http://www.turtles.org/identify.htm>). The front flippers of sea turtles form semi-rigid hydrofoils (Walker 1971; Font et al. 2011), which provide nearly flat areas that allow the use of mathematical algorithms that match planar surfaces. The National Advisory Committee for Aeronautics (NACA) 0014 airfoil has been used to model the turtle’s front flipper (Font et al. 2011). Posterior of the maximum thickness the profile provides a nearly flat surface. The proposed method has been used when areas of the flipper are obscured while the turtle is resting in and under coral. Matches might also be possible on the uninjured portions of an injured flipper, which should match similarly to partially visible flippers. To the author’s knowledge this is the first systematic in-water Photo-ID study using sea turtle flippers.

Photos of juvenile green sea turtles were obtained using Canon SX280HS digital cameras (12MP) with a WP-DC42 waterproof housing while snorkeling and SCUBA diving during one to three week annual summer trips to Bonaire from 2012 to 2016. Bonaire is an island in the Caribbean Sea approximately 80 km from Venezuela. Details on sea turtle density, nesting and other information can be found in the STCB (Sea Turtle Conservation Bonaire) annual reports (<http://www.bonaireturtles.org/wp/explore/publications>). A set of 64 images of left flippers of 30 individual green turtles were selected from the dive sites Karpata, Andrea I, Cliff, and Salt Pier. Left front flippers were used for no other reason than the author had more left front flipper pictures than right front flippers. Images captured while SCUBA diving and snorkeling with different illumination, viewing angles and quality were selected to test the capabilities of the prototype application. Included in the author’s dataset are eleven turtles encountered only once, eleven photographed intra-year and eight in multiple years.

I³S Classic (den Hartog & Reijns 2014) was explored for computer assisted flipper matching because of its successful prior use with sea turtle facial matching. The user manual advises viewing angles of no more than 30°. The program requires three reference points that are visible in all pictures and their exact location must be consistently and clearly distinguishable. The author could find no such landmarks defined for sea turtle flippers. Usable reference points were discovered after self-study of sea turtle flipper photographs and consulting a sea turtle flipper anatomical diagram (<http://www.fairmanstudios.com/turtle-flipper-anatomy>). The reference points selected are the first scale intersection proximal of digit one, first scale intersection proximal to where digit two would emerge and the most proximal intersection of the scale covering digit five (Fig 1). Other spot points, which are typically scale intersection points, also need to be picked which can consume some time and may be inconsistent even when chosen by a single individual. The author abandoned I³S Classic because the three reference points are not always visible, may be missing in the case of an injured turtle and the time and difficulty involved in consistently picking spot points.

Figure 1. Left front flipper imager of “Hook” showing (A) outlined digit scales and (B) reference points.

Figure 2. (A) Bonaire green turtle on 13 August 2014, courtesy Erik van der Zijden; (B) author’s photo;(C) flipper from 2A affine transformed in to flipper from 2B. (C) Affine mapping of flippers from (A), colored red, and (B), colored green, and yellow indicates overlap.

Figure 3. Illustration of simple grouping based on number of scales posterior from the digit two scale. (A) One scale and a partial coded at 1.5 (B) two scales and a partial as 2.5 (C) Three scales as 3.0 (D) Three scales and a partial as 3.5.

Because the three reference points define a plane, the author employed ImageJ with a geometric mapping plugin (<http://ij.ms3d.de/geometric_mappings.php>) that provides an affine transform. Verification of both the usefulness of these reference points with different flipper poses could be matched was tested. Two images of the same turtle on a single dive taken by two divers with different cameras, angles and lighting show the efficacy of this method (Fig. 2).

The areas surrounding the reference point may also be used to classify flippers into groups similar to head scales (Schofield et al. 2008; Lloyd et al. 2012). Figs. 3B, 3C and 3D have been affine transformed using the reference points to match the Fig. 3A view. The author uses these classifications when manual matching. Table 1 contains the digit two pattern number which is coded as the number of scales directly posterior to the digit two reference scale. Partial scales are usually encoded as an additional 0.5 (see Fig. 3). The pattern of scales around digit five has also been used by the author for quickly confirming or rejecting a match while manually searching images and verifying weak computer-generated matches. Reference points were only used to verify that affine transformations are a suitable method of matching flippers and are not used for the prototype application.

Having established that affine transformations can allow for matching flippers, Affine Scale-Invariant Feature Transformation (ASIFT) (Yu & Morel 2011) was tested as a tool to match sea turtle flipper images. ASIFT leverages Scale-Invariant Feature Transform (SIFT) (Lowe 2004) and is designed to match images with much wider viewing angle differences than SIFT. This aids in matching in-water photos where the viewing angle of the turtle cannot be controlled. ASIFT models six viewing parameters instead of four for SIFT. SIFT has been used previously for Photo-ID of leatherback turtles (Buonantony 2008) and manta rays (Town et al. 2013).

Table 1. Details and matching results of author-provided turtle flipper dataset.

Figure 4. Visual matching results for green turtle image pair. Convex hull match area shown on image (A) that has been transformed to view of (B).

A prototype application based on ASIFT was built by the author. Reference implementations of ASIFT can be found in OpenCV (Bradski 2000) and Yu & Morel (2011). Frames from a movie of a green turtle in “flight” near Little Cayman compared to a reference image using the prototyped program and recombined into a movie (<http://www.youtube.com/watch?v=ubcwmCG1ZWM>) illustrates that flippers may be matched successfully in a variety of poses during active swimming and validates that regions of the flipper can be treated as planar surfaces. Unlike other methods (den Hartog & Reijns 2014; Dunbar et al. 2014), no manual point/intersection picking is required. The preprocessing of images consisted of rectangular cropping the flipper from the raw images and optionally adjusting brightness/contrast and rotation using ImageJ (Schneider et al. 2012). Better images are taken against a plain background such as blue water or white sand. This can eliminate the false matching of similar objects in the background of both images. Rotation before cropping can eliminate much of the background if needed. Using the author’s dataset the prototype application scales large images reducing computer resources. Other modifications were made to the algorithm to reduce computer resources because ASIFT is typically 13.5 times the computational effort of SIFT (Yu & Morel 2011). The cropped images may be of different sizes so the matching process is repeated with the image order swapped. The match score used by the author is the sum of matches in both directions. The minimum match scores for all images of an individual are reported in the Min Match Score column of Table 1. In all true match cases the minimum match scores were greater than any non-matching image scores, which are provided in Table 1 as the Max non-match Score column. The highest scoring non-match turtle name is also included in Table 1 as column Non-match turtle. Turtles appearing several times as non-matching indicate that they have at least a small region that is similar to several other turtles as in the case of “Flyer” which has the highest false match score for five turtles. A visual example of the prototype application is illustrated by Fig. 4. Fig. 4A is a cropped image of Fig. 5A that has been transformed by the application into the view of 4B which is a cropped image of Fig.5B. The application, when run interactively, draws the convex hull of the matching points on the first image to highlight the area where matches were made. This aids the user in verifying whether the image pair is a weak match or a false positive. The application has been in use by the author for several years to validate uniqueness and stability of flipper scale patterns over time. Table 2 details 183 encounters with 34 turtles near the Red Slave and Salt Pier dive sites that were successfully Photo-IDed multiple times for periods up to 1,477 days. Thirteen turtles were tracked for more than 1,000 days. One turtle “J” was recaptured after 1,333 days. Turtle “OC” has been followed via Photo-ID from before getting an external tag (Fig. 6A), after tagging (Fig. 6B) and finally after tag loss (Fig. 6C). Salt Pier is an industrial and heavily used recreational site so this data may be useful for studying the health and population dynamics of the resident turtle population.

Figure 5. Original uncropped images for results shown in Figure 4. (A) “Flyer” on 01 August 2016 night snorkel. (B) “Flyer” on 23 August 2014 day snorkel.

Table 2. Turtle encounters verified by prototype photo-ID application. Site abbreviations: RS = Red Slave SP = Salt Pier.

Further research could explore other scoring methods and criteria for matching key points as well as algorithm optimization. Vacationing and resident snorkelers and divers can be leveraged to acquire useful Photo-ID images as part of their normal aquatic activities while following local conservation laws and guidelines. Fig. 5B shows a nearly ideal image while snorkeling clearly showing both flippers and includes both the carapace and the top of the head that can also be useful in identification. Photos used in Photo-ID can be taken from a distance to minimize barrel, perspective and other possible distortions in the image. Turtle flipper scale patterns can be used as a low cost, non-invasive matching technique while turtles are actively swimming, resting or foraging. The author suggests this method can be applied to the juvenile stage turtles of other species that have visible scale flipper patterns. As an example, Fig. 7 shows the technique applied to a hawksbill turtle (Eretmochelys imbricata) recaptured after 1,868 days. In that time period the turtle was tagged. Photo-ID extends the turtle’s recorded life history. From the author’s volunteer work photographing captive juvenile loggerhead turtles (Caretta caretta) could also benefit from this technique. The reference points and groupings may be useful for both hawksbill and loggerhead turtles. The method has been shown to be useful for periods up to 1,477 days for Chelonia mydas and 1,868 days for Eretmochelys imbricata.

The source code for the prototype application is available for non-commercial use from the author. Fig. 8 shows an example of the prototype application when run in interactive mode. The standard ASIFT algorithm can be accessed on the Image Processing On Line website (<http://ipol.im>).

Figure 6. Turtle “FC” tag history: (A) 01 August 2017, (B) 30 December 2019, (C) 28 May 2020.

Figure 7. Hawksbill turtle. (A) 14 August 2015 affine transformed to (B) 24 September 2020 (Courtesy David Whillock).

Figure 8. Application display. Matching images upper and lower right. Detail of match points from bottom right image on left.

Acknowledgements. The author thanks the following individuals and organizations for assistance with this project: Sharon Pursley (dive partner) for her time commitment and assistance in conducting turtle observations; Anya and Honey Pursley for brainstorming sessions during walks; Zsuzsanna Pusztai, Dedrie Pedersen for passing on their love of sea turtles; Ben Higgins and NOAA Fisheries sea turtle facility staff for turtle education and animal interaction; Erik van der Zijden for a photograph; Cathleen and David Whillock for photographs; Julia Reisser and Maíra Proietti for turtle head image dataset; Sue Willis and Sea Turtle Conservation Bonaire (STCB) staff and volunteers; Coral Paradise Resort, Dive Friends Bonaire, Captain Don’s Habitat, Sea Sports SCUBA and Sheila Shelton for dive support; and Peter and Ursula Bennett for pioneering Photo-ID of sea turtles. The author also thanks the reviewers of this study.

ARAUJO, G., J. MONTGOMERY, K. PAHANG, J. LABAJA, R. MURRAY & A PONZO. 2016. Using minimally invasive techniques to determine green sea turtle Chelonia mydas life- history parameters. Journal of Experimental Marine Biology and Ecology 483: 25-30.

BENDIK, N., T. MORRISON, A. GLUESENKAMP, M. SANDERS & L. O’DONNELL. 2013. Computer-assisted photo identification outperforms visible implant elastomers in an endangered salamander, Eurycea tonkawae. PLoS ONE 8: e59424.

BENNETT, P., U. KEUPER-BENNETT & G.H. BALAZS. 2000. Photographic evidence for the regression of fibropapilloma afflicting green turtles at Honokawai, Maui, in the Hawaiian Islands. In: Kalb, H. & T. Wibbels (Comps.). Proceedings of the 19th Annual Symposium of Sea Turtle Biology and Conservation. NOAA Tech Memo NMFS-SEFSC-443. pp. 37-39

BENNETT, P. & U. KEUPER-BENNETT. 2001. The use of subjective patterns in green turtle profiles to find matches in an image database. Proceedings of the 21st Annual Symposium on Sea Turtle Biology and Conservation. 2005. Coyne, M. & R.D. Clark (Comps.). NOAA Tech Memo NMFS-SEFSC-528. pp. 115-116.

BOLGER, D.T., T.A. MORRISON, B. VANCE, D. LEE, D. & H. FARID. 2013. A computer-assisted system for photographic mark recapture analysis. Methods in Ecology and Evolution 3: 813-822.

BRADSKI, G. 2000. The opencv library. Doctor Dobbs Journal 25 11: 120-126.

BUONANTONY, D. 2008. An analysis of utilizing the leatherback’s pineal spot for photo-identification. M.E.M. thesis. Nicholas School of the Environment and Earth Sciences, Duke University, Durham, North Carolina, USA. 50pp.

CAILLOUET, C.W., JR., D.B. REVERA, M.J. DURONSLET & J. BRUCKS. 1989. Dermatoglyphic patterns on Kemp’s Ridley sea turtle flippers: Can they be used to identify individuals? Proceedings of the 1st International Symposium on Kemp’s Ridley Sea Turtle Biology Conservation and Management. Texas SeaGrant Report TAMU-SG-89-105. pp.146-150.

CARPENTIER, A.S., C. JEAN, M. BARRET, A. CHASSAGNEUX & S. CICCIONE. 2016. Stability of facial scale patterns on green sea turtles Chelonia mydas over time: A validation for the use of a photo-identification method. Journal of Experimental Marine Biology and Ecology 476: 15-21.

CHASSAGNEUX, A., C. JEAN, J. BOURJEA & S. CICCIONE. 2013. Unraveling behavioral patterns of foraging hawksbill and green turtles using photoidentification. Marine Turtle Newsletter 137:1-5.

DE URIOSTE, J.A., M.J. BETHENCOURT & H. SICILIA. 2016. Sea turtle photo-identification. In: Rguez-Baron, J.M., M.M. Lara-Uc & R. Riosmena-Rodriguez (Eds.). Advances in Research Techniques for the Study of Sea Turtles. Nova Science Publishers, Hauppauge, NY. pp. 53-70.

DEN HARTOG, J. & R. REIJNS. 2014. I³S Classic Manual. Interactive Individual Identification System. Version 4.0.2 <http://reijns.com/wp-content/uploads/2020/01/I3S%20Classic.pdf>

DUNBAR, S.G., H.E. ITO, K. BAHJRI, S. DEHOM & L. SALINAS. 2014. Recognition of juvenile hawksbills Eretmochelys imbricata through face scale digitization and automated searching. Endangered Species Research 26: 137-146.

FONT, D., M. TRESANCHEZ, C. SIEGENTAHLER, T. PALLEJÀ, M. TEIXIDÓ, C. PRADALIER & J. PALACIN. 2011. Design and implementation of a biomimetic turtle hydrofoil for an autonomous underwater vehicle. Sensors 11: 11168-11187.

GATTO, C.R., A. ROTGER, N.J. ROBINSON & P. SANTIDRIÁN TOMILLO. 2018. A novel method for photo-identification of sea turtles using scale patterns on the front flippers. Journal of Experimental Marine Biology and Ecology 506: 18-24.

HALL, A.G. & J.B. MCNEILL. 2013. Inferring sea turtle recapture rates using photographic identification. Herpetological Review 44: 561-569.

JEAN, C., S. CICCIONE, E. TALMA, K. BALLORAIN & J. BOURJEA. 2010. Photo-identification method for green and hawksbill turtles - First results from Reunion. Indian Ocean Turtle Newsletter 11: 8-13.

LLOYD, J.R., M.A. MALDONADO & R. STAFFORD. 2012. Methods of developing user-friendly keys to identify green sea turtles (Chelonia mydas L.) from photographs. International Journal of Zoology 2012: 317568.

LOWE, D.G. 2004. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision 60: 91-110.

MORRISON, T.A. & D.T. BOLGER. 2014. Connectivity and bottlenecks in a migratory wildebeest Connochaetes taurinus population. Oryx 48: 613-621.

REISSER, J., M. PROIETTI, P. KINAS & I. SAZIMA. 2008. Photographic identification of sea turtles: method description and validation, with an estimation of tag loss. Endangered Species Research 5: 73-82.

SCHNEIDER, C.A., W.S. RASBAND & K.W. ELICEIRI. 2012. NIH Image to ImageJ: 25 years of image analysis. Nature Methods 9: 671-675.

SCHOFIELD, G., K.A. KATSELIDIS, P. DIMOPOULOS & J.D. PANTIS. 2008. Investigating the viability of photo-identification as an objective tool to study endangered sea turtle populations. Journal of Experimental Marine Biology and Ecology 360: 103-108.

SU, C.M., C.T. HUANG & I.J. CHENG. 2015. Applying a fast, effective and reliable photographic identification system for green turtles in the waters near Luichiu Island, Taiwan. Journal of Experimental Marine Biology and Ecology 467: 115-120.

TOWN, C., A. MARSHALL & N. SETHASATHIEN. 2013. Manta Matcher: automated photographic identification of manta rays using keypoint features. Ecology and Evolution 3: 1902-1914.

VALDÉS, Y.A., J.A. RICARDO, F.B. TRELLES & O.E. ABAD. 2014. First assay of photo-identification in marine turtles’ nesting population. Revista de Investigaciones Marinas 34: 43-51.

WALKER, W.F., JR. 1971. Swimming in sea turtles of the Family Cheloniidae. Copeia 1971: 229-233.

YU, G. & J. MOREL. 2011 ASIFT: An algorithm for fully affine invariant comparison. Image Processing On Line 1: 11-38.