No. 1 Story

ACCC clears Optus to scrap HFC network and use NBN instead

The ACCC has cleared, provisionally, the proposed deal between Optus and NBN Co under which Optus is to be paid around $800m to shut down its HFC network and transfer customers onto the NBN. read more

Related Articles

Adoption of cloud computing has reached a tipping point  - but don’t expect legacy...
In yet another blow to the Facebook IPO this week, following the withdrawal of...
Recruitment technology and social media have played a significant role in growing business in...
Want a free trip to E3 Expo? Microsoft's looking for an 'Xbox Insider.' Microsoft Australia...
Online DVD rental and Internet movie distributer, Quickflix (ASX: QFX) has announced a partnership...

Google Image Swirl puts a spin on image search

Your IT - Home IT

Google's experimental Image Swirl puts a new spin on finding similar images. A visual interface clusters similar images.
Google Image Search as long been a handy way of tracking down pictures of particular people, places or things. The recent addition of "Find similar images" made it easier to zoom in on exactly what you're looking for.

The latest improvement is Google Labs' Image Swirl. In essence, it combines the Wonder Wheel feature that presents similar concepts (you'll find that by clicking the Show Options link at the top of a Google search results page) with image search results.

An initial Image Swirl search returns a dozen stacks of related images. Click on the one that seems closest to what you're after, and the original results swirl to one side, with the larger part of the window being taken up with the contents of the selected stack arranged in a hub-and-spoke format.

Click on one of the newly revealed images and the results swirl again as you drill down through the successive layers.

imageswirl.jpg
















It's important to remember that Image Swirl is just a new way of presenting similar images. Whether images really are similar depends of Google's algorithms, which sometimes give strange results.

For example, the results of a Google Image Swirl search for "cup" include a photo of a yellow-green cup and saucer. One of the linked images is a map of Canada, presumably because the colour is vaguely similar and Hudson Bay provides a topological equivalent of the cup's handle.

And you can be taken quickly down some strange paths. That same cup is seemed to be similar to a Mango Butter Cup (a soap-like cosmetic product) pictured in front of its box, which is deemed similar to a paper cup making machine, which in turn is likened to a rugby ground.

Part of the problem seems to be reliance on the words in the filenames or surrounding text rather than just analysing the images themselves.

Still, the way Image Swirl clusters generally similar images does show potential for finding just the right image more quickly. It would be even more useful if it incorporated Image Search's ability to include usage rights as a search criterion, even if that feature isn't 100% reliable.