Advances in machine vision may soon allow you to search using images instead of keywords.
That may not seem wholly intuitive, but images contain a wealth of information, which makes searching via user-uploaded images a tool with lots of potential. Engineers at Shutterstock, which has a library of 80 million stock photos, have been experimenting with machine vision and machine learning to augment their search capability.
Right now when you search for an image on Shutterstock, algorithms match your search with related keywords attached to images in the library. But keyword data, while pretty useful for indexing images into categories, isn't as effective for surfacing the best and most relevant content. So Shutterstock's computer vision team worked to apply machine learning techniques to reimagine and rebuild that process.
The result is pretty cool, and an early look at a new way of searching that's going to become a lot more relevant as machine learning and machine vision improve.
Shutterstock's search technology now relies instead on pixel data within images.
According to VP of engineering Kevin Lester: "It has studied our 70 million images and 4 million video clips, broken them down into their principal features, and now recognizes what's inside each and every image, including shapes, colors, and the smallest of details; this visual and conceptual data is represented numerically."
That means you can search for images that have the look and feel of another image you particularly like, which expands the dimensions of searchable qualities in Shutterstock's library significantly. Instead of searching for "San Francisco," for example, you can now search for particularly grainy photos with a certain kind of lighting.
Search applications utilizing machine vision and machine learning are bound to expand. One day soon you'll be able to point your smartphone camera at anything in the visible world and retrieve relevant information based on image search technology.