Pattern recognition in cryptic wildlife species is an evolving field resulting from the increasing use of photographic images resulting from non-invasive techniques such as infra-red, “heat-in-motion” detecting cameras in the field.  We would like to introduce this concept, explain the problem and solicit ideas from the scientific community on techniques and possible solutions.

 

Concept: Tools used in the field of pattern recognition can be applied to the study of cryptic wildlife species using “camera-trap” capture-recapture techniques. 

 

Photographs of margay (Leopardus wiedii, see photo), for example, form a digital “fingerprint” of the individual, and with photos of both sides of a cat, we can identify this individual each time it passes a camera.  Work thus far also shows promise that the tails of these cats might have the most symmetrical markings. 

 

Currently methods for inferring absolute density estimates from wildlife populations using cryptic patterns for individual ID require two cameras – one for each side of the individual as they are not symmetrical from side to side.  Current image analysis techniques involve simply looking at photos from each side of an animal and creating a visual “signature” for that individual. However in this case two camera-traps are required at each site in order to get paired photos.

 

Because camera systems are relatively costly and access to remote field sites is difficult we are exploring the idea that individuals can be identified from non-paired images of the left and right side of the animal along the back and top of the tail.

 

Problems and solutions:

Technical Problem:

The surface on which cryptic patterns exist on cats, i.e. the fur, is susceptible to changes in distance and shape contortion between spots and markings as the animal moves.  Even a simple walk in front of the camera seldom, if ever, gives identical signatures, so we tend to focus in on specific spots or regions which are hopefully always characteristic of that individual.

 

Markings across the back and tail have some symmetry, and from this observation, we are looking for ways to match the sides together—to map (if you will) the pattern with two halves.  It would seem that a flexible image tool could be developed to match (with some error of course) two independent photos of different sides of the same individual.  This could be used to positively identify this individual with a defined statistical probability.  We imagine a method that can give us a nice little statistical readout, such as “image A and image B have an 85% with a probability of matching”. Having done this by hand it seems feasible, but cutting out photos of cat torsos and tails and matching them by eye hardly seems a viable solution.

 

To save time and money we are exploring the possibility of using just one camera at each study site and hope to develop a technique for analyzing this “single-side” data by matching patterns across the back and tail.

 

Possible solution:

Image rectification: contorting the black and white pattern of the animals to a neutral by warping the image.

 

Other challenges: Photos do not always contain the entire animal (i.e.,  sometimes just the torso and tail, other times just a head and torso) and can be from a variety of vantage points (i.e., animal facing camera etc.), thus complicating the pattern to be recognized.

 

In long-term studies however a collection of photographs from the same individual can be used to put the pieces together and create a “signature” or “fingerprint” for that side of the individual, but matching left to right sides still remains problematic.

 

Measures of success:

Measuring the statistical probability of positive identification of individuals varies with photo quality, whether or not both cameras fired, the proportion of animal in the image (among numerous other factors) and is difficult to determine. Higher sample sizes, greater image interpretation experience and trap location and placement will certainly influence this probability. The question is whether this probability would significantly change if cameras are not paired. Our feeling based on preliminary analysis is that it is possible with either some interdisciplinary collaboration with specialist in this field or training of wildlife biologists. We, the wildlife scientists, would like to develop a technique for cryptic species “fur-print” analysis and possible develop and integrated software tool to improve out research and conservation of wild cats and other cryptic wildlife species.

 

We are working on this problem in the field in the Talamanca region of Costa Rica using images of jaguar, ocelot and margay. We invite any suggestions or collaborations from interested students or professional and are willing to share all image data and publications.

 

Can you help?

Pattern Recognition in Cryptic Wildlife Species

 

By:  Jan Schipper

schi6037@uidaho.edu

leopardos@earthlink.net