theInfor


Sketch Image Search

Posted in Uncategorized by theinfor on the January 10, 2006

Retrievr LOGO
Sometimes when I was doing bits of image design work, I always want to search some images for further work as rough materials according to the contents. I have clear mind myself about what these images that I need should be like, such as a green eye or a people standing by certain gesture. Of course, I can “Google” it with terms that can represent the images characteristic the most. However, the fact is publishers of those images don’t always name the picture in similar ways; a facetious photographer could possibly name a picture of a beautiful female’s eye as “am I a sexy man in that little heaven?” Undoubtedly, Google wouldn’t send that picture back to you no matter what term you are using except ones like “heaven”.
So, as an alternative for searching images without considering what names of the images are, the sketch image search become available online from retrievr. It let users to draw a rough sketch on the drawing panel they provide, and then perform a matching between what the user just draw and what it has according to its index. Currently the index is mainly based on some of Flickr’s images. And as announced on the introduction page, the application is still an experimental service. However, it can work very well if the rough sketch supplied by users is drawn properly. The most important two factors that impact application’s performance are colour definition and shape of the sketch. I actually made some experiments, some of the results turned out to be very satisfying. For example, by just drawing a Cambridge blue dot on the panel, seconds later, I get several beautiful blue eyes shining on the page, whatever the images names are! I also draw a red heart-shaped pattern with little black dots added on it, and an image of strawberry came out just as expected!
But there are limits of the application. Here is some introduction from the site:

One thing to keep in mind is that retrievr doesn’t do object/face/text recognition of any kind, so if you’re drawing an outline sketch of a chair, it almost certainly won’t get you one back (except your index only contains images of chairs). The same holds for corporate logos, icons &c.
It helps to think of it as matching the most pronounced shapes and slabs of colors.
Another thing to know is that there’s currently no way to specify the aspect ratio, so you have to rescale the image in your head (things that are close to the borders of the image you’re thinking of should be close to the borders of your sketches), but that’s really more of a missing feature of the drawing flashlet than an inherent problem. Sometimes it also helps to remove detail instead of adding it.
Personally, I see retrievr more as an “exploration” tool than as a “search” tool, and it seems to work very well for that.

Anyway, the publication of this site may mean a milestone in image search history. I hope the technology can rapidly grow up.

Leave a Reply