As Google continues to build out its search technology, the company has gradually expanded beyond browser-based text search to include voice and image searches.
The experimental
Google Goggles is one such project: an image-based search service that allows people to submit an image and get Web-based search results pulled from Google's massive archive of indexed photos and images. It's currently available only on Android-powered smartphones, but a Google developer is working on a side project to incorporate the same functionality into future versions of the company's Chrome Web browser.
The concept is simple enough: snap an image with your phone, and let Google Goggles search the Web for results based on the image you've submitted. But it's not always easy to execute.
I put Google Goggles to the test on my Motorola Droid running Android 2.0.1. Logos and text seem to be the simplest tasks for Goggles to handle. It had no problem identifying an older version of the CNET logo, returning results with an image of the logo as well as our Web site, cnet.com. Other tests didn't go quite as well.
Here, you can see one image I uploaded, next to the results Goggles returned based on that image. Kirkwood is a ski resort near Lake Tahoe. Its logo is similar to CNET's in that it includes both text and the company logo. But oddly enough, Google Goggles translated this image as a bar code with the number 15882135, and pointed me toward results featuring Franklin, Ind.
Outside of CNET's office in downtown San Francisco, I tried out the live view augmented reality features of Google Goggles, which are intended to overlay information on shops and restaurants in front of you based on the GPS location of the device. You don't need to take a picture to use this feature, instead simply use the live camera view through Goggles.
With my Droid's GPS turned on, Goggles failed to give very accurate results nearby. Standing outside Tara, a Thai restaurant on Second Street in San Francisco, Goggles was unable to identify my location and instead returned results for restaurants a block or two away.
Even snapping an image of another restaurant storefront logo that included a text name did not accurately return results for the Imperial Tea Court, a tea store inside San Francisco's Ferry Building.
In its oddest display of information during my walk, I snapped an image of my Canon 5D Mark II on the ground to see if it would return accurate search results for the camera.
Instead, Goggles returned to me results for Van Huesen Factory Store, which was located across the street from my location.
At the Ferry Building, a prominent landmark at the end of Market Street, I snapped an image that included this lighting fixture, which seemed to confuse the search. It returned no results.
Moving to a more unobstructed view, Goggles was able to correctly identify the Ferry Building, pointing me to its Web site, FerryBuildingMarketplace.com as the first result, and the Wikipedia entry as the second result.
Product search is another use for Goggles, and in my tests it worked very well. There are many mobile applications that allow you to scan a bar code for results, but Google Goggles simply needs an image of the product, like this bottle of wine from the Napa Valley in California.
Goggles accurately returned an image result for the wine bottle label, as well as a link to a review of the specific 2006 merlot.
Books were another successful search for Goggles. Even when flanked on either side by information I was not searching for, Goggles was able to find David Sedaris' book "When You Are Engulfed in Flames," offering to compare prices at Google Product Search, as well as offering to preview the book at Google Book Search.
Another product I tested Goggles on, the cult favorite Sriracha hot sauce, successfully returned results for an image of the product as well as the Wikipedia entry for it.
This piece of street art along Howard Street in San Francisco includes a few illustrations I thought Goggles might recognize. Goggles successfully returned results for one of the illustrations, Abraham Lincoln, while ignoring the Bay Bridge, Sutro Tower, Tom Selleck, and the Ferrari 308 GTS from the TV show "Magnum P.I."
The 24-70mm 2.8 lens on my Canon camera contains wraparound text, so I thought it might pose a challenge to Goggles. Goggles was able to identify it successfully, returning results for similar Nikon, Olympus, and Sigma 70mm zoom lenses.
All in all, Google Goggles was not perfect, but was still impressive. Users of the Android app have the choice of enabling or disabling submission of their image search history to Google, and I would expect that the more the feature gets used, the better it will become.
