I use image search often, mainly with Google. I use it to find inspiration in design or better clipart, and occasionally as a content searching feature. If I’m looking to find info about a new Fender guitar or Corvette, I might use the image search; then once located, I may continue on to the content related to that picture. That content, after all, is the main ranking factor for images.
If you use image search, you probably already realized that it, well, sucks eggs compared to text search. The first handful of results are often in the ballpark, but it quickly goes extremely south from there. Unlike text search, where a few websites full of content can often suffice, in image a much larger set of results is usually desired. I can almost guarantee that without the ‘adult content’ filter on, you’re going to get some kind of completely unrelated nude picture a third of the time. When using image search at work, be careful that nobody is standing behind you. Those pictures can be freaky!
Obviously, it’s hard to favor image search and really back it if it’s so wonky.
But the future of image search always sounded interesting. Improvements on the idea of engines using apps to map the parts of a picture, determining the shapes (including faces), and using that as part of algorithm sounds pretty cool. Imagine, no matter what the image file was named, or the content around the image file on websites, by image searching for “left arrow clipart” or “Scarlett Johansson”, you’d actually get accurate results. This recognition technology is used in government, but seldom used online (a great exception is the soft goods comparison shopping engine Like.com).
Bill Slawski, search patent extraordinaire, recently posted about MSN’s new patent filing on an advance to their image search. Check out that article! Below are some of Bill’s factor highlights:
Expanding on what’s happening now, engines may be able to go deeper than on-page keywords. “It might also look for text that is associated with an image, but is found on a different pages (perhaps links to the picture, and possibly text associated with those links).”
The algorithm may begin to recognize types of queries, like a person’s name, and set forth a new set of detection methods to only serve pics of people.
But how about these factors?
- Number of Websites that Contain an Identical or Similar Image
- Quality of the Image (related to size of the image to user value)
- Link Relationships Between Images (i.e. thumbnail images linking to relevant pages/images)
- Frequency of an Image Within a Website (the more the image appears, the less ’special’ it is)
Some cool insight into how engineers think, and some hope for the future of image search. Maybe not as ‘futuristic’ as it could be, but cool nonetheless.
May 4th, 2008 at 4:22 pm
nice article. i thought image search was sort of an added search that nobody would really use much, but this makes sense. i can probably think of some other good ways to search for things with image search maybe.
May 5th, 2008 at 12:25 pm
Here’s a link I found from Search Engine Journal related to Google’s possible future with image search. This IS more along the lines of what I expect out of image search.
May 15th, 2008 at 10:53 pm
This tells me that search engines are still really pretty young in the grand scheme of things.