We all use Google to look for images. Normally you type in a keyword and you’re given millions of pictures that match it. But how often do you use a reverse image search?
A reverse image search is what you do when you already have the image in hand, but want to find out more about it. Generally you upload the image to the search engine, or type in the URL of the image you have in mind. Then the search engine checks its database and shows you everywhere on the web that your image is posted. Not too shabby.
Reverse image searches are nothing new, but they’ve been improving a lot (my personal favorite is TinEye). They can often depict variants on your target image, such as larger or smaller versions, different res, or even a larger image that yours has been cropped from.
Wolfram Alpha, however, is doing something different. Their new search engine, ImageIdentify, does not find pictures similar to yours—it tells you what’s in your picture.
In other words, upload a picture of a cat and it says it’s a cat; upload a picture of a hot air balloon and, you guessed it, it will know that too. The accuracy is not great yet, but that’s no big deal because it will improve rapidly with time. In fact, the site encourages you to grade accuracy each time you search, and will learn from your corrections (one of the grading options is, “What the heck?!”).
Cool as it is though, I do have to wonder: what’s the point? How often do you have a picture of something and not know what it is? It seems like this would only help for the most obscure and confusing images, or if it gave very specific answers like which species of spider is in a photo. It’ll be a long time before any image recognition technology can do that.
But it will be possible someday, and I think that’s what Wolfram is really up to. They’re building the tech and letting it learn from experience. That lays the groundwork for future machines that have almost an almost supernatural ability to identify things from even the tiniest clues.
So go ahead and check it out. It’s fun uploading pictures and seeing just how right (or wrong) Wolfram can get them, and every time you type in a correction you’re helping build the future of smart machines.