Here are two truths:
1.Some people are always looking for easier ways to do hard things.
2.Technology can be amazing.
There are a lot of people anxiously waiting for WeBIRD, the promised birdsong equivalent to Shazam and MusicID, but they may have to wait a little bit longer. Music databases can make a match to a digitally produced song, but they can’t match you singing exactly the same song. The variations in the human voice make that kind of analysis a less than exact science. A person can recognize that you are singing Happy Birthday; Shazam can’t. Similarly, there is enough variation in birdsong that it’s very difficult to get a computer to recognize the nuances. You can teach one to recognize some of the songs some of the time, but a reliable tool is not available yet.
But can we use technology to recognize visual cues and help us to ID birds? New birders are being encouraged by some to post their birding photos to the Internet in order to get an ID, instead of using field guides or birding mentors. Crowd-sourcing identification is certainly one way to handle it, but it would be much “cleaner” if we could somehow get the computer to do the work for us, wouldn't it? Enter Google Search by Image. Seriously! You can upload an image or provide a URL to images.google.com, and Google will search for similar images. How cool is that! Just click on the little camera at the right of the search field and follow the instructions.
When I learned of this, I knew an experiment was in order! I uploaded a picture of a Hermit Thrush, clicked “Search” and waited to see how Google would handle the difficult Catharus species challenge. My uploaded image shows at the top of the screenshot below:
Can you say “Epic Fail”? None of these pictures is even close, except for the background colour tones. None were even of birds, let alone thrushes. I guess I should have expected that. Maybe it was much too challenging. How about something simpler, like a Barred Owl? The Internet is crawling with owl pictures. This should be easy!
Sigh… Not even a little bit better. More diversity in the selection of “matches”, but still no birds and certainly no owls. Maybe the whole bird has to be visible. Here’s a Burrowing Owl, Google. What can you do with this?
Um, no. But it’s interesting to see how many celebrities resemble Burrowing Owls. Brangelina? The algorithm seemed to be focussing on color-matching. What about a bird with a distinctive color and shape? Easy--Great Blue Heron!
Eureka! It matched one! Admittedly it’s the fifth image the program chose, and it somehow thought that a better match for my heron was a staged suicide scene ( in the top row), but at least it got a bird, and the right bird at that!
I was prepared to completely dismiss this function as useless, but then an interesting thing happened. A birder from Ontario sent me a picture he took while visiting Vancouver to see the Red-flanked Bluetail. It was a great photo of a bird that he (and those I showed it to) identified as a Veery, an almost unimaginable bird to be in Vancouver this time of year. But with a Brambling and a Bluetail around, never say never, right?
People send me pictures all the time, but there was something about this report that made me suspicious. Spidey-sense, some people call it. I asked for more information, which did not come. I did a little online detective work and didn’t find anything reassuring. So I posted the report--along with my reservations--on the Vancouver birding bulletin boards, mindful of those who think that all rare birds should be reported and not wanting anyone to miss out on this potential rarity. Then I remembered the Google Search by Image tool. I uploaded my suspect image--another Catharus--and guess what? Here are the results:
Epic win! The person reporting this bird was a prankster (very funny-not!), my spidey-sense was on the mark, and within seconds, Google found the image in an almost-two-year-old blog post from Massachusetts. I don’t know what motivated the hoax, but I’m delighted that the perpetrator was found out before an onslaught of inevitably frustrated birders wasted their time.
For bird ID, Google Search by Image has a very long way to go. There are some things that humans can still do better than our current technology. But today, I, and all my lookalikes below, are giving a big alula up to Google!
I'm not sure of the details, but the Cornell Lab of Ornithology is apparently "teaming up with computer vision researchers at UC-San Diego, UC-Berkeley and Caltech to tackle the challenge of using computers to identify birds in images."
http://dev.nabirds.org/Web/Tools/ImageUpload/pages/about.php
Posted by: Corey Husic | 02/12/2013 at 08:32 AM