image matching algorithms

[Daniel Fetchinson]

> > The photos are just coming straight from my digital camera. Same
> > format (JPEG), varying size (6-10 megapixel) and I would like to be
> > able to pick one and then query the database for similar ones. For
> > example: I pick a photo which is more or less a portrait of someone,
> > the query should return other photos with more or less portraits. If I
> > pick a landscape with lot of green and a mountain the query should
> > result in other nature (mostly green) photos. Something along these
> > lines, of course the matches won't be perfect because I'm looking for
> > a simple algorithm, but something along these lines.
> >
>
> Ah. In that case, SIFT isn't for you. SIFT would work well if you have
> multiple photos of the same object. Say, a building from different
> angles, or the a vase against different backdrops.
>
> If I'm understanding your correctly, what you're attempting here is very
> general and well into the highly experimental. I've been wishing for
> such a feature to appear in something like Google Image Search
> (pick/submit a photo and return similar images found on the web). I'm
> sure if there's even a practical solution, Google (or MS) would be on it
> already.
>
> The problem is that there isn't really one. Despite what you may see
> claimed in university press releases and research papers, the current
> crop of algorithms don't work very well, at least according to my
> understanding and discussion with researchers in this field. The glowing
> results tend to be from tests done under ideal conditions and there's no
> real practical and commercial solution.
>
> If you restrict the domain somewhat, there are some solutions, but none
> trivial. You are probably aware of the face searches available on Google
> and Live.
>
> The histogram approach suggested by Shane Geiger may work for some cases
> and in fact would work very well for identical resized images. I doubt
> it will work for the general case. A mountain with a grassy plain at
> noon has quite a different histogram from one at sunset, and yet both
> have related content. Manual tagging of the images, a la Flickr, would
> probably be your best bet.

Since you seem to know quite a bit about this topic, what is your
opinion on the apparently 'generic' algorithm described here:
http://grail.cs.washington.edu/projects/query/ ?
So far it seems to me that it does what I'm asking for, it does even
more because it can take a hand drawn sample image and query the
database for similar photos.

There is even a python implementation for it here:
http://members.tripod.com/~edcjones/pycode.html

On the histogram method I agree that it won't work partly because of
what you say and partly because it is terribly slow since it's
comparing every single pixel.

Thanks,
Daniel