Look for images of “Washington” online, and Google’s search engine may turn up a random sea of pictures showing the Washington monument, the White House, George Washington, and actor Denzel Washington. Now Google’s new “Image Swirl” feature could eliminate that hit-or-miss frustration by organizing images in neat, expandable thumbnail stacks for users to explore.
Google launched its new Google Labs feature today as a more exploratory and interactive alternative to the usual image search. Click on a thumbnail stack of the Washington monument, and the image cluster expands outward in a swirling pattern to form a constellation of sub-cluster images. Those sub-clusters might organize themselves by characteristics such as night or day views and different lighting angles, based on Google algorithms that crawl across the Internet looking for similar visual cues, meta tags, text descriptions and even faces.
People who know what specific picture they want can already make use of Google’s “Similar Image” feature — assuming they find the right image among the swarm of search results. By contrast, the new Image Swirl really shines as a tool that provides the bird’s eye view for people who don’t know what they are looking for.
“We think it’s particularly important for visual information, because that’s more about exploring rather than finding,” said Aparna Chennapragada, Google product manager for Image Swirl. “We think the interface should allow for that.”
We’ve previously reported on Goog’s in-house project focused on sorting landmark images in online photo collections. The same types of visual algorithms used in that project help power Image Swirl, which represents one of the first consumer products to emerge from Google’s computer vision research.
But Image Swirl also incorporates better algorithms that organize images by a broader set of criteria, said Yushi “Kevin” Jing, a Google software engineer. Whereas the landmark image project could identify 50,000 unique landmarks this past summer, the new Image Swirl feature can filter 200,000 of the most popular online image queries — and each query consists of many image sub-clusters.
Curious parties can provide feedback to help Google decide whether to incorporate the feature in its main search engine. Google even provides suggested fill-in queries to nudge users toward one of the existing 200,000 active queries. We at PopSci plan to take Chennapragada’s advice and give “Apple” a try, among other queries.