Searching For a Better Image
Ask Jeeves has launched a brand-new image search service, using some innovative techniques to identify high-quality pictures on the web. How does it stack up against the competition?
Ask Jeeves has launched a brand-new image search service, using some innovative techniques to identify high-quality pictures on the web. How does it stack up against the competition?
Ask Jeeves has launched a brand-new image search service, using some innovative techniques to identify high-quality pictures on the web. How does it stack up against the competition?
Image search has been gaining popularity among web searchers. Indeed, the Keynote Systems survey I wrote about last week said that 63% of Yahoo users use the specialized image search function from time-to-time. In launching its new image search service, Ask Jeeves says that 16% of all searches it handles are in the image search area.
Jeeves has had a picture search feature since mid-2003, but it has been powered by Picsearch, a third-party provider of image search capabilities. Ask Jeeves’ new image search was developed internally and uses many of the features that make its web search feature unique, in addition to some new technologies specifically developed to find imagery.
Image search is tricky, because images lack most of the clues search engines use to find relevant text documents that match our queries. Because images are made up of patterns of bits rather than words, search engines can’t directly “look” at an image and figure out what it represents.
Instead, search engines look for other clues, such as filenames, text immediately above or below an image (potential captions), the overall context of a page an image appears on, and so on. Ask Jeeves image search is doing all this, but it’s also applying its Teoma ranking system to find sites that have a broad representation of images and topics, and identify those that have the greatest degree of “authority” for a particular topic to help determine image relevance.
The company also says that it’s applying sophisticated image recognition technologies, examining factors such as color, brightness and other characteristics to better understand what an image represents.
How’s the quality? Well, by nature image search is much more subjective than text-based search. An image that appeals to me may look like total garbage to you, and vice versa. In my tests, I found that I generally really liked the image search results provided by Ask Jeeves for most of my queries, such as patrick o’brian, gunflint trail and search engine strategies (amazing how a picture of the Ask Jeeves booth turned up in the first page of results!).
The coolest feature of the new image search feature is the addition of the Zoom related search results. The Zoom query refinement feature that Jeeves added to web search results last year offers additional search suggestions, but rather than presenting a list of related topics that are essentially keyword driven, Zoom presents suggestions that are conceptually categorized into three segments: Narrow Your Search, Expand Your Search and Related Names.
This makes it easy to find pictures related to those you’re currently viewing. For example, a search for Steve Jobs brings up Zoom links for the related names Steve Wozniak (Apple’s co-founder), Bill Gates and others. Interestingly, these related names go beyond contemporary computer people—there are also links to Charles Babbage (inventor of the computer precursor called a “difference generator,” Ada Lovelace (the first programmer of the aforementioned machine) and other notables in the history of computing.
How does the new image search stack up to others on the web? In the blog post announcing the new image search service, James Jeude, Senior Product Manager invites us to try three different queries and see how they compare across the various engines.
For the query bondi beach (Ask Jeeves Pictures, Google Images, MSN Search, Yahoo Images) all services returned interesting images of the Sydney beach. I couldn’t honestly say that one was better than any other—they all had a good mix of results.
The query rickenbacker (Ask Jeeves Pictures, Google Images, MSN Search, Yahoo Images) proved more interesting, as it could refer either to the famous World War I flying ace or to a guitar. Jeude rightly points out that Jeeves’ returned images of both, but claims “Our competitors: there is no variety of topic.”
While it’s true that neither Google nor Yahoo had a picture of the red baron on the first page of results, they did feature images of him on subsequent pages. And MSN image search results (powered by Picsearch) had just as much of a variety as Ask Jeeves.
With the query washington (Ask Jeeves Pictures, Google Images, MSN Search, Yahoo Images) Ask Jeeves trumped Google and Yahoo hands-down, displaying images of the man, the state and the monument. Once again, MSN did a good job (it also included a picture of actor Denzel Washington).
How does Ask Jeeves image search compare with a user-tagged collection such as Flickr’s (bondi beach, rickenbacker, washington)? Subjectively, you get a very different set of pictures from Flickr than you get from any of the search engines, especially when you sort by “most interesting” rather than recent pictures. You also get far fewer results, which in some cases can be a good thing.
The bottom line: Ask Jeeves new image search is a step ahead in a notoriously tricky area. With the quality of its image search results, combined with the new Zoom query refinement feature, I’ll be using it as my default image search service going forward. That said, I fully expect to continue using all of the image search services described, given the relatively undeveloped state of the technology in general.
NOTE: Article links often change. In case of a bad link, use the publication’s search facility, which most have, and search for the headline.