Idee - full of ideas

By Martin Joergensen | July 13, 2008 5:30 PM | Permalink | Comments ( 1)

multicolr.jpgI have been following the company idée for a while and had fun with their fascinating TinEye search system. The system requires an invite to use it, but it seems to be pretty easy to get one. I registered and have been using the image search a few times since.

You need to try it to really understand it, but the essence is that you can search the web for images similar to one you choose or upload. There's an extension to FireFox, which allows you to just right-click an image and launch a search on TinEye. The thing is that TinEye uses an image-comparison algorithm to match pictures. Not filenames, tags or anything bland like that. This means that TinEye will find scaled images, cropped images and images altered in other ways - even retouched. It even has a way of comparing them right in the search results. Amazing. But the people behind idee don't stop there. Go to their lab page and sample some of their wild ways to work with images. Their multicolr search engine is really fascinating. It matches 3 million flickr-images live as you select colors from a swatch. Select 1-2-3-4-5 up to 10 distinct colors and multicolr will almost instantly select a set of images that perfectly match these colors. It's simple and really amazing - but certainly not trivial seen with a programmer's eyes. I program myself and have fooled around a bit with this type of algorithm, but I have been nowhere near the level and power of a facility like this.

Along the same lines you will find BYO Image Search, which matches one of your images or an image online to a selection of Alamy-images using texture, composition and color. It does a really stunning job, and my trials were very convincing.

The reason that I find idée's ideas interesting is that with the ever increasing amount of images available - online as well as on our own hard drives - new ways of ordering and searching them are necessary. We cannot rely on tagging and keywords when the numbers grow to tens of thousands, millions and even billions of images. The machines have to help us, and techniques like idée's are some of the ways to go.



Try a very professional image search engine which gets similar images by many categories such as shape, color and visual rank.
It also search in all the web using yahoo. flickr or google's picasa and let's you get images with a creative common license for commercial use.