Research designed to examine the behavior and ecology of cetaceans has benefited greatly from the ability to recognize individuals (International Whaling Commission 1990). The use of distinctive, naturally occurring variations in the appearance of individual cetaceans within a population has been commonly combined with photo-identification techniques (Wrsig and Jefferson 1991). Bottlenose dolphins (Tursiops truncatus) are particularly well suited to the process of photo-identification (Wrsig and Wrsig 1977). The thin posterior edge of the dorsal fin becomes "notched" during interactions with conspecifics, predators, and humans, often resulting in recognizable patterns of nicks, scars, and notches. Such patterns are analogous to human fingerprints and are unique to each individual. Notch patterns once acquired by a dolphin are usually permanent with little change occurring over time.
The Marine Mammal Research Program (MMRP) of Texas A&M University at Galveston has been conducting studies on bottlenose dolphins along the Texas coastline since 1990. Research funding has been provided by grants to Dr. Wrsig from the National Marine Fisheries Service, Sea Grant College System and the Marine Mammal Commission. The MMRP has now identified over 1,500 individual bottlenose dolphins from Texas coastal waters, and this number is expected to approach 2,500 once data analysis is complete. Members of the MMRP have also assembled identification catalogs of bottlenose dolphins from Argentina, Costa Rica, California and Baja, Mexico totaling well over 3,000 individuals. As the number of dolphins identified within a population increases, the physical process of "matching" individuals becomes labor intensive, and the probability of making errors increases significantly. Each new photograph obtained in the field must be compared to all previously cataloged photographs, a task requiring substantial effort and skill. Recent advances in computer-based image recognition software hold promise for developing the requisite technology from which a computer-assisted dorsal fin matching program can be created.
We propose that a synergistic effort between cetacean biologists at the MMRP and image recognition specialists of Texas A&M University Galveston be funded to develop and implement this needed and important biological tool. Specific research objectives are twofold: 1) to create an efficient and accurate image recognition program capable of individual bottlenose dolphin identification from dorsal fin photographs; and 2) develop an associated digital storage and retrieval system for large photo-identification collections. The proposed research will not only transform the methodological technique of photo-identification but will also significantly contribute to monitoring efforts of cetacean populations worldwide, ultimately allowing realistic management and conservation policies to be created.
Research Personnel
Dr. Phillip Levin - Professor of Marine Biology at Texas A&M University Galveston and Director of the Image Analysis Laboratory. Levin has expertise in statistical methods applied to the dynamics of populations of marine organisms (Levin 1994, Levin and Hay 1995), as well as in the use of image analyusis as a tool to facilitate an understanding of marine populations (Levin 1994). Over the last eight years, Levin has developed considerable expertise using image analysis systems similar to the one to be used in this study for analyzing and classifying photographic and microscopic images of protozoa, plants, invertebrates, and vertebrates. In particular, Levin has used image analysis to analyze: the microstructure of fish otoliths (earstones), algal community structure on temperate and tropical reefs, morphometric data from various invertebrates, and to enumerate positions of parasites on their hosts. Levin will be the Principal Investigator and will supervise the project.
Dr. Bernd Wrsig - Professor of Marine Biology at Texas A&M University, Professor of Wildlife and Fisheries Sciences at Texas A&M University and Director of the Marine Mammal Research Program. Qualifications include: development of the photoidentification method as a technique for the study of bottlenose dolphins; 25 years of field research on the behavior and ecology marine mammals worldwide; expertise in photography and photo analysis. Wrsig will act as an non-paid consultant and serve as the cetacean specialist responsible for developing the conceptual framework for the empirical testing of the image recognition program.
Dr. William E. Evans - Professor of Marine Biology at Texas A&M University, Professor of Wildlife and Fisheries Sciences at Texas A&M University, President of Texas Institute of Oceanography and Director of the Bioacoustics Laboratory. Evans has extensive experience in marine mammal research including the application of new technology to the study of cetaceans and the use of photo-identifiation of color patterns and natural markings in Killer whales and common dolphins (Evans 1994, Evans and Yablakov 1983). He will act as a non-paid advisor.
Diane Blackwood - Doctoral Student at Texas A&M University, Department of Wildlife and Fisheries Sciences. Qualifications include: experience in hardware and software design; creation of automated feature recognition in jaw motion and EMG analysis (Blackwood 1991, Tate et al 1994); expertise in signal processing techniques, digital signal processing, and production programming. Blackwood will design the analysis techniques and write the software necessary to implement the automatic dorsal fin matching.
Wesley R. Elsberry - Doctoral Student at Texas A&M University, Department of Wildlife and Fisheries Sciences. Qualifications include; experience in hardware and software design; expertise in artificial intelligence and artificial neural network techniques applicable to pattern recognition; and production programming, and software engineering (Elsberry 1989). Elsberry will design the analysis techniques and write the software necessary to implement the automatic dorsal fin matching.
David Weller - Doctoral Student at Texas A&M University, Department of Wildlife and Fisheries Sciences. Qualifications include; 12 years of research experience on the cognition, behavior and ecology of bottlenose dolphins; development of photoidentification methodology and photographic cataloging systems (Weller 1991, Defran et al. 1990); management and analysis of longitudinal photo-identification datasets from Texas, California and Baja, Mexico. Mr. Weller will serve as an expert advisor on photoidentification techniques and principles. He will supervise undergraduate processing of slides including sorting, selection, scanning and image preparation. He will hand select critical points to act as a model expert for the computer to emulate.
Current Methods
Dorsal fin photographs are taken from small research vessels with 35 mm cameras using color slide film or black and white film. Laboratory photo-analysis begins with the initial sorting and identification of a collection of negatives from a photographic survey. Only high quality photographs of distinctive dorsal fin notch patterns are included for analysis. These quality photographs or "type specimens" are then sorted into discrete individual files. Type specimens are then rear projected and enlarged so that the dorsal fin notch pattern may be traced in a 10-cm x 17-cm frame drawn on white paper. This tracing process results in uniform hand drawn replications (tracing) of each individual identified in the initial sorting process. A dorsal ratio is then calculated for all fins with 2 or more notches (Defran et al. 1990, Weller 1991). The dorsal ratio is the distance between the two largest notches on the fin, divided by the distance from the lower notch to the top of the dorsal fin. As a relative measure the dorsal ratio is unaffected by the size of the fin when photographed, enlarged, or even under moderate cases of parallax, yielding a scale invariant parameter. Once calculated, the dorsal ratio is recorded directly on the tracing, and used to facilitate matching to previous photographs already existing with the database. If a tracing cannot be matched in its appropriate catalog, then all tracings in all catalogs are inspected twice. Although labor intensive, this systematic search process ensures that all previously sighted dorsal fin notch patterns, including those which may have changed, will be resighted. If the tracing is not matched after thorough inspection of all catalogs, then the individual is considered a new sighting.
Proposed Methods
Computer assisted matching will be based upon the fundamental methods currently in use at the MMRP (described above). The automation of this task is anticipated to significantly reduce labor and potential errors made while matching. A multi-phase approach will be used in developing and implementing such an automated system. This approach begins with two basic problems which need to be solved. One is automatic detection of distinctive features of a dorsal fin. The other is then using these distinguishing features to match new fins to those in the pre-existing library. Much of the work will be accomplished using macro programming techniques within the Optimas image analysis system. The Optimas software is already in place at the Image Analysis Laboratory. Additional pattern recognition and automation work will be done in the C programming language. The following is a descriptive framework and time-line for accomplishments:
Design Phase - (2 months) This phase will explore which features of dolphin dorsal fins are most important in identification. Both the reliability of each feature and the frequency of its presence will be examined. During this time, a software system to compare bitmapped images of dorsal fin photographs will be designed. Files will be treated to have all dorsal fins mapped to the same color and algorithmic comparisons will help to create points of similarity beyond those already determined by the pre-existing dorsal ratio system.
Phase One - (2 months) A test group of negatives will be scanned with distinguishing points marked by photo-identification experts. These critical points, such as dorsal fin tip, top and bottom of each notch, and the deepest point of each notch, will also include other points identified during design discussions. A preliminary classification system will be developed based on these data points. All classification systems will include normalization of data before comparisons are made.
Phase Two - (2 months) Software feature detection methods will be used to locate critical points automatically. The scanned negatives in phase two will still require the use of imaging software to define what part of the image is dolphin and what part of the image is non-dolphin. Once the dorsal fin silhouette is defined, boundary detection software will be used to delimit the outline. Then, two-dimensional signal processing techniques and feature detection methods will be used to select points that correspond to those selected in phase one.
Phase Three - (3 months) Features of the dorsal fin outline that cannot be easily selected by human experts will be examined. This will include measures of curvature for both the leading and trailing edge of the dorsal fin and selection of the base of the fin to aid in normalization. These features along with those described in phase two will be used to further separate dorsal fins for identification.
Phase Four - (2 months) Difference functions between a stereotypical 'ideal' dorsal fin and the dorsal fins found on negatives will be implemented.
Evaluation - (2 months) After implementation of phases 1-4, an evaluation of the most effective classification and matching cues will be made. At this time analysis with our photo-identification domain experts will re-evaluate user needs required for the final product.
Phase Five - (6 months) Design and implementation of a dorsal fin classification and matching protocol will begin. Factors learned from the first four phases will be used to design this tool, and will be re-written as a new program at this time. Forcing a re-write at this point should result in cleaner and more efficient program code with a friendlier user interface (Brooks 1975).
Testing - (4 months) The final protocol will be tested by software and photo-identification domain experts, and additional streamlining and refinements will be made by programmers as needed.
Institutional Commitment and Sources of Additional Support
This study has the full sponsorship from the Marine Mammal Research Program, Image Analysis Laboratory and the Center for Bioacoustics of Texas A&M University at Galveston. The Marine Mammal Research Program houses an extensive library of bottlenose dolphin dorsal fin negatives, photographs and slides including over 80,000 images representing 1500+ individuals. Data from this library will be available for design, testing and use of the image recognition program. Computer facilities including Optimas software, Pentium computers, and a Polaroid slide scanner will be available through the Image Analysis Laboratory. The Center for Bioacoustics will supply additional Macintosh computer support.
Impact on Infrastructure of Science and Engineering
This research will provide support for three graduate students over the period of the study and will also provide partial support of undergraduate workers.
The development of an accurate and efficient computer-assisted photo-identification system will significantly impact marine mammal studies world-wide. While the technique of photo-identification has been in use since the early 1970's, this time-tested biological research technique has yet to be fully interfaced with the available computer science technologies capable of improving, streamling and refining the method. Photo-identification studies on bottlenose dolphins are particularly valuable because this species is a top level predator which often inhabits the same nearshore waters as those frequented by humans. Thus, bottlenose dolphins serve in many respects as important biological indicators of the status and relative health of nearshore habitats. Studies using photo recognition of individuals have been critical to our understanding of this species' distribution, numbers, residency patterns, and range characteristics. Thus, it is critical that this technique is improved and updated to not only enhance our understanding of marine mammal populations but to also monitor the habitat upon which they are dependent.
Budget and Justification
Slide Scanner $3,000
Mass storage (CD writer) $10,000
Graduate Student Stipends $64,800
Undergraduate workers $6,000
Software $1,000
Miscellaneous supplies $1,000
Domestic Travel $500
Total $86,300
Since the photographic images are 35mm slides, a slide scanner is needed for digitization. The use of one slide scanner will be provided by the Image Analysis Laboratory. A second scanner is needed due to the large volume of slides to be processed. Mass storage will be required to hold the library of over five years of photographic data. For long-term use, recordable CD-ROM provides a cost effective means of storing digital data. The budget includes the cost of a CDR writer, a non-thermal compensating hard disk for CD-ROM image formation, and enough CDR blanks to permit testing of the pilot system over the period of the study. Graduate student stipends are needed for partial support of Diane Blackwood, David Weller, and Wesley Elsberry over the period of the study. Undergraduate students will assist in preparing and scanning photo-ID slides into a digital database. Qualified undergraduates may also work as assistant programmers under the supervision of Elsberry and Blackwood. The purchase of Adobe Photoshop software will enable the rotation of images to a standard orientation. Miscellaneous computer supplies will include such expendable supplies as computer paper, toner cartridges, and floppy disks. Domestic travel is requested for one person to attend a scientific meeting for the purpose of presenting the results of this research.
Bibliography
Blackwood, D.J. 1991. Analysis of Motion and EMG During Mastication in Subcondylar Fracture Patients. Master's thesis, University of Texas at Arlington & University of Texas Southwestern Medical Center at Dallas.
Brooks, F.P. 1975. The mythical man-month : essays on software engineering. Reading, Mass. : Addison-Wesley Pub. Co.
Defran, R.H., Shultz, G.M. and Weller, D.W. 1990. A technique for the photographic identification and cataloging of dorsal fins of the bottlenose dolphin (Tursiops truncatus). In: Individual Recognition of Cetaceans: Use of Photo-Identification and Other Techniques to Estimate Population Parameters, P.S. Hammond, S.A. Mizroch, and G. P. Donavan (eds). Reports of the International Whaling Commission; Special Issue 12:53-55
Elsberry. W.R.. 1989. Integration and hybridization in neural network modelling. Master's thesis, University of Texas at Arlington.
Evans, W.E. 1994. Common Dolphins (Delphinus delphis limaens). (in) Handbook of Marine Mammals, Volume 5, (eds) S. H. Ridgeway and R. Hanson. Academic Press. Cambridge.
Evans, W.E. and Yablokov. A. V. 1983. Variability in cetaceans color patterns: anew approach to the study of mammalian coloration. Nauba Publishers. Moscow. (in Russian, English Abstract). International Whaling Commission 1990. Individual recognition of cetaceans: Use of Photo-identification and other techniques to estimate population parameters. Reports of the International Whaling Commission (Special Issue 12), Cambridge
Levin, P. S. 1994. Recruitment in a temperate fish: the roles of macrophytes and food supply. Env. Biol. Fsh. 40:271-281
Levin, P. S. and Mathieson, A. C. 1991. Variation in a host-epiphyte relationship along a wave exposure gradient. Mar. Ecol. Prog. Ser. 77:271-278.
Tate, G. S., Throckmorton, G. S., Ellis, E., III, Sinn, D. P., and Blackwood, D. J. 1994. Estimated masticatory forces in patients prior to orthognathic surgery. J. Oral Maxillofacial Surgery. 52:130-136.
Weller, D.W. 1991. The social ecology of Pacific coast bottlenose dolphins. Master's Thesis, San Diego State University, CA. 93 pp Wrsig, B. and Wrsig, M. 1977. The photographic determination of group size, composition, and stability of coastal porpoises (Tursiops truncatus). Science 198:755-6
Wrsig, B. and Jefferson, T. A. 1990. Methods of photo-identification for small cetaceans. In: Individual Recognition of Cetaceans: Use of Photo-Identification and Other Techniques to Estimate Population Parameters, P.S. Hammond, S.A. Mizroch, and G. P. Donavan (eds). Reports of the International Whaling Commission; Special Issue 12:43-52