AI isn’t all about getting machines to generate university essays. Indeed, its impact on the visual world is likely to create even more headlines. In the world of search, image recognition is taking huge strides forwards. For many searches on Google, we’re now as likely to be presented with images as we are text links, if the result is gauged to be more appropriate. Try searching for “welcome to scotland road sign” and you’ll get photos of the sign, not a page describing it.
But how did Google identify those images? Traditionally, it would have been reliant on captions and ‘alt’ text on the image. Later, it would learn how to reliably work out what an image contained by the text on a page around it. More recently, when images contained text, it would apply image and character recognition to the text, something the search above probably takes advantage of. Now, however, it’s undoubtedly employing the kind of sophisticated machine vision we’ve been developing in industry for years.
Help with recognition
What does this mean for us in terms of search engine optimisation? Of course, the standard captions or ‘alt’ text remain important. But to help with image recognition, there’s little to lose by choosing to use photos which are ‘clean’, and can be identified more easily. In other words, single products, on a plain background, shot from conventional angles.
If image recognition algorithms look at a series of pages about blue widgets on different sites, and see a number of images with similar shapes, they’ll learn the outline of a blue widget. If we have our product photographed in an odd way, or on a background of confusing objects, it’s going to make the AI’s task to make the connection a lot harder.
I’m not suggesting that we re-shoot our most attractive photos, or choose a bland product shot instead of an in-situ one for application articles. But if there is a choice, and it’s unlikely that it will make much difference to the sales story, using the photo that a machine should prefer might be the tiebreaker.