Microsoft researchers have been testing a new way of searching for pictures. The system, which they call ImageFlow, shoots images at the user in 3-D, like bullets. Yesterday, the software giant released a paper on its pilot test.
Indeed, image search services such as Google Images are in some need of innovation. Results are typically displayed in a 2-D grid, in order of relevance. But people don’t search for images the way they search for text. Research has shown that image searches differ in three ways: They are more likely to be conducted for entertainment, there’s more of an emphasis on resource gathering, and there’s a higher rate of query reformulation–trying to be more precise about what you’re looking for, in other words–than there is with text searching.
ImageFlow aims to improve the experience by adding an extra dimension. Once you enter a search query in a box at the top of the screen, thumbnail images begin to fly at you. “The desired effect is that of users continuously ‘flying’ though a star field of images, where the depth of an image is directly mapped to its relevance,” explain the designers in their paper. Images further away in the star field are less relevant–they would appear lower down in the traditional grid.
You can take command of your starship in several ways. You can control your front-and-back flight through space with a mouse wheel, or by clicking a scrubber on the left side of the screen. You can also move up and down, or left and right–but these motions affect something other than relevance. In the test set up by the Microsoft team, images along the color-to-black-and-white scale were mapped along the y-axis. Along the x-axis were related search queries, in an effort to accommodate the query refiners. In the example given by the researchers, a search for “Michael Jackson” would first turn up flying images of the King of Pop. Then, if you drag the canvas left or right, you’ll activate a new, related search query: “as one moves to the right,” write the authors, “the result images will show images of Michael and of Janet Jackson until all of the images correspond to the search ‘Janet Jackson.'” They put together a little color coded diagram to illustrate this.
Finally, since many of us are hunting for multiple images, the researchers developed a “basket” that can be expanded and minimized, where you can collect images you want to hold on to.
The results of the experiment were somewhat mixed. The research paper notes that the 12 volunteers for the new system reported that it was “different,” “cool” and “fun,” but also that it was “distracting.” Flying through a star field is a nice way to browse, but it doesn’t let you see enough images at once to help you find what you need. And just because image searchers often reformulate their queries doesn’t mean that they reformulate it to the next-most-searched-for query. Who’s to say that when I search for pictures of Michael Jackson, the next thing I want to search for are images of Janet?
Still, the idea of experimenting with image search is a good one–and one that Microsoft’s competitors have been working on for a while. If the description above has whet you’re appetite for a new kind of image search, try out Google’s “image swirl,” a labs feature first launched in 2009 that organizes images by category and presents them in circular clusters.