ad1

Breaking News UPDATED AT 1:21 PM EST Monday, Dec. 8, 2003
Tech Home | e-insider | Reviews | @Play | Today's Paper | Investor

Wonderfile sees value in visual photo searches


By Sue Bowness
Globe and Mail Update

E-mail this Article E-mail this Article
Print this Article Print this Article   

  • Company: Wonderfile, a royalty-free stock photography Web site
  • Vendor: Idée, provider of visual search technologies
  • Product: Espion, a visual search solution that allows users to analyze, index and search large sets of images and video based on visual content
  • Project: Incorporating Espion visual search technology into Wonderfile's established keyword search process as "simsearch"



    Visual searching technology is one of those ideas that has to be seen in order to be fully appreciated.

    For Wonderfile (www.wonderfile.com), a royalty-free stock photography agency with hundreds of thousands of photos to search through, the benefits of visual searching are obvious. John McDonald, vice-president of marketing for Wonderfile, describes the challenge: "When you do a keyword search, the good news is that you might get 30,000 images; the bad news is also that you might get 30,000 images."

    The problem with developing a search engine for images, as Leila Boujnane, chief executive officer of Idée http://www.ideeinc.com/, points out, is this: "Put three people in front of a painting and they will all describe it differently."

    The same idea applies when labelling images, since different photo archivists may label a field of flowers as "field' or "flowers' or "sunflowers" or even "summertime" or "happiness."

    Rather than forcing the reader to page through these results one by one, visual searching allows the user to narrow their search by relying on visual similarities rather than highly subjective keywording.

    Say the user wants to find a photo of a field of flowers, probably sunflowers, but they're open to other types. The user starts an initial search with by entering the keyword "flowers," which produces approximately 3,000 images from the Wonderfile database. Rather than scrolling through pages of photos containing images of everything from single tulips to rose bouquets to fields of wildflowers, the viewer can choose a picture that looks similar to what they're looking for — say, a field of yellow flowers - and click on 'simsearch' below the image. The chosen image is then treated as the original, compared with other images in the collection for visual similarities such as colour, layout and other properties.

    The images in the collection are instantly rated in terms of accuracy relative to the "new original", so after only a couple of simsearches the user can arrive at a page with fewer than 100 images, most of which are sunflowers. Suddenly the user is exactly within the range they're looking for, and more likely to buy something on the site.

    As Paul Bloore, chief technology officer of Idée explains, "What the visual search allows [the user] to do is go through a few sets and then narrow it down and say 'this is what I'm looking for'. And when they find it, they will purchase it. That's what we see as the benefit."

    You don't have to be a photo researcher to realize the significance of this application for stock photo agencies, archives, newspapers, or even industries like film and gaming where they need to keep track of huge image pools or video clips. In the stock photo world, besides the human time saved in rooting out duplicate images from photo collections — Ms. Boujnane estimates that at least seven- to 15-per-cent of a typical stock collection is duplicates — the technology allows creative departments to track image modifications like cropping and colour changes. In addition to initially searching by keyword, the user can also start with initial image that they can scan in, or by drawing the image themselves in a paint program.

    With visual similarities driving the search process, the user can also find images that are unexpected. For example, the user might have entered 'sunflower' because it came to mind as a cheerful yellow flower, but through visual searching might discover that an image of a field of brown-eyed susans is closer to their actual imaging needs.

    By allowing the number of potential images to be narrowed more quickly, users are freer to let their sense of creativity guide them, says Mr. McDonald, "the client drives the site, the technology helps you find pictures according to your instincts."

    With implementation of the entire system in under four weeks, Idée's customization also went smoothly for Wonderfile, the company said. According to Mr. McDonald, the most difficult aspect was devising an intuitive interface.

    "On a Web site people do not read - so the hardest part was to get an interface that seems reasonable," he said.

    Mr. McDonald describes the technology as having "paid for itself already in terms of the licensing fee."

    Simply visiting to the Wonderfile site, even the most unsophisticated user can see why simsearch is a valuable tool, and why Mr. McDonald doesn't even hesitate when he says that, "visual search will be an absolute necessity in the future."







Tech Poll
Do you subscribe to any e-mail news alert services?
Yes
No

Results & Past Polls
globeinvestor.com globeandmail.com ROBTv Workopolis CTV.ca TSN.ca Discovery.ca
© 2003 Bell Globemedia Publishing Inc. All Rights Reserved.
Bell Globemedia