I don’t mean the bird species, which is found in North America and also in different varieties elsewhere in the world. I’m not talking about the wizard, either. I’m referring, rather, to the Merlin Bird ID app that I wrote about last month. It turns out that Cornell Tech and Caltech, working together as a team called Visipedia, have been developing a new tool with the Cornell Lab of Ornithology for Merlin Bird ID that involves computerized identification of bird photos. Called Merlin Photo ID, this beta-stage program can take an image of one of North America’s most common bird species (a pool of 400) and identify it after a human user has pointed out where its bill, eye, and tail are.
And after testing it out for a bit I learned that it doesn’t even need all of those data points! I started out with an easy shot of a Red-winged Blackbird. The program asks where and when you took the photo, so that it can use the same eBird data the normal Merlin Bird ID app uses to figure out the most likely species. Then it asks you to box in the bird and click on the points you can see in my screenshot below:
A male Red-winged Blackbird should have been a quick ID for the program, and it was! Next I tried a male Indigo Bunting, also a fairly simple matter for the artificial intelligence. Then I gave it a slightly trickier shot of an Eastern Phoebe, which has six “similar species” listed in its All About Bird page by the Cornell Lab of O. Phoebes and other flycatchers are often identified with the help of distinct calls and behavior, and obviously the Merlin program doesn’t have the benefit of these field notes, but nevertheless it distinguished my photo from the potentially confusing Willow Flycatcher or Eastern Wood-Pewee — both of which, I’ll point out, I saw the same day as the pictured phoebe!
Finally, I decided to see if a missing data point would mess with Merlin Photo ID to the point that it wouldn’t put the correct bird as the most likely species. I gave it a photo of a Great Crested Flycatcher, admittedly not a terribly hard bird to identify if you’re not in a range border zone, where the whole tail and leg region was blocked by a bush, so the program only had the eye and bill points to go by. After a slightly longer calculation time than the previous attempts (and still well under a minute), Merlin Photo ID got it right again! Eventually this automated identification tool will get incorporated into the current Merlin app, but even while limited to computer web-browser use at the moment I’m sure lots of people will find it quite helpful.