Turtle Identification


WHAT ARE WE DOING?

The Center for Biodiversity and Conservation at the American Museum of Natural History (CBC-AMNH) has been studying green and hawkbill sea turtles on Palmyra Atoll in the Central Pacific. To help understand turtle movements we are developing methods for using the scale patterns on a turtle’s head to identify and recognize individual animals. Even though computing power continues to rapidly increase, humans still have a huge advantage over computers when it comes to recognizing patterns associated with details in images. Here we are trying a very simple experiment to see how well people can identify turtles with minimal training, and comparing these results with other matching methods. By crowdsourcing the effort, we can tap into a large group of people interested in contributing to the study.

Using a test dataset consisting of photographs of 102 individual green sea turtles, some taken over multiple years, we have 48,985 image pairs that need to be compared to determine if it is the same animal in both images. Each image pair will be randomly shown to participants 10 times over the experiment for statistical rigor. This means we have a total of 489,850 comparisons that need to be made! Once all image pairs have been matched, the results will be compared to 2 other methods we are evaluating: a purely automated process based on computer vision and machine learning algorithms and to a hybrid method involving both human input and computer algorithms.

WHY ARE WE DOING IT?

To understand turtle movements within a region or ocean basin, external flipper and internal Passive Integrated Transponder (PIT) tags are commonly used. We are developing photographic methods for using the scale patterns on a turtle’s head to identify and recognize individuals. Each individual’s scales form a unique pattern – similar to a fingerprint. A photographic approach has a number of advantages: it is less costly than other tagging methods, involves less stress to the animals, and any photo taken in the turtle’s natural habitats, in known locations, can contribute valuable information. This way, many people from around a region can collect and contribute to studies similar to ours. The more we know about where turtles are and how they are moving around, the better we can better manage and conserve their populations and habitats.

HOW CAN I HELP?

You can help by looking through pairs of images and determining if the same animal appears in both! Once you launch the matching interface you will see two images. You can use your mouse wheel or the icons below the image to zoom in and out. Click and drag to pan around the image once you have zoomed in.

Turtle Matching Interface

Using the shape of the scales and their pigmentation, look closely to determine if the same animal appears in both images. Note, pigmentation can change a little overtime or be obscured by algae on the turtle. Matches are rare and the quality and orientation of the animal in the image can really complicate the determination too, so look closely, carefully, and don't rush! We need you to submit your best determination, even when it seems the image pairs are too difficult to compare. 

Below is an example of a very challenging match, i.e., the same animal does appear in both images.

GD11_1

GD11_2

The orientation of the turtle's head is very different in each image. These images were taken several years apart and there are slight changes to the pigmentation on the scales. Oh yeah ... and there is a GIANT water spot in one of the images.

Not all images pairs are as challenging as the one above but notice once again the slight pigmentation differences that can exist between years?

GC4_1

GC4_2

 

 
I am ready to start matching!