When you search for something, it pulls the first 50 image results from Bing Image Search then blends them all together. The resulting image is produced by taking the mean color value of each pixel across all of the images.
This is prime territory for PCA and eigenimages
<a href="http://en.m.wikipedia.org/wiki/Eigenface" rel="nofollow">http://en.m.wikipedia.org/wiki/Eigenface</a>
This is (surprisingly) very cool! Some of these results could easily be hung on a wall.<p>I'd love it if you could add some sort of crude voting system, so that the good ones rise to the top.
So, just pull and average. No attempt to align, no attempt to decompose, no attempt at PCA, no processing at all. Just pull and average.<p>I'm struggling to see why this is interesting.<p><i>Added in edit: Rather than just downvoting, perhaps you could tell my why this is interesting. Were there technical challenges to overcome? If so - what? What did the implementor learn by doing this? What are you learning by using it? Please - help me to see why this is at all interesting! I genuinely don't understand.</i>
This is a really cool project, good job!<p>Spoon is an interesting example : <a href="http://imgessence.com/browse/view/190" rel="nofollow">http://imgessence.com/browse/view/190</a> - You can see the many basic outlines of a 'spoony' shape. However, all of those images of spoon are presenting the object rotated at some more or less random angles. It would be interesting to add some algorithm that would try to match the images (rotate / scale) to the average and hence give less noisy output.
These are the best ones I've managed to create so far:<p>Circle: <a href="http://imgessence.com/browse/view/329" rel="nofollow">http://imgessence.com/browse/view/329</a><p>Benzene: <a href="http://imgessence.com/browse/view/303" rel="nofollow">http://imgessence.com/browse/view/303</a><p>Nike logo: <a href="http://imgessence.com/browse/view/299" rel="nofollow">http://imgessence.com/browse/view/299</a>