The Semantic Web is Easy: Let Computers Do It
Ha.
Even if we could automagically classify, categorize and connect all the textual content on the web, what would we do with the 2 billion + photos on Flickr? Well, ALIPR (that’s Automatic Linguistic Indexing of Pictures in Real-time) is going to take care of all that tricky classification for us by using image recognition (and some other stuff) to automatically tag all of our photos.
It is not hard to guess how this works right now. But even if the software improves, good 202ers know that the problem is far from solved.
Sunny Lee Said,
October 16, 2008 @ 12:12 pm
I decided just to check it out and see how it works. I uploaded an image of a puppy I found on dailypuppy and the following were the tags that showed up: plant, bird, wild_life, flower, rural_england, lake, grass, landscape, everglade, animal, building, rural_France, house, historical, garden. haha. Clearly, conceptually sound but needs some work. Also it took me 3 trials to get a photo to successfully upload. The site kept telling me that my images were too small or something. shrug.
Ryan Greenberg Said,
October 19, 2008 @ 11:35 pm
The best and most useful recognition software I have used is WhatTheFont. It lets you upload an image file of text and it identifies what typeface is in the image. Although it is quite accurate, it does suggest the long road to great generalized image recognition. In order for WhatTheFont to work well it needs an image with good resolution and black on white type. The procession conditions are tightly constrained because WhatTheFont knows what to expect: letters. ALIPR, on the other hand, can expect any book from the Library of Babel. Currently it seems to be drawing on a relatively limited vocabulary. I suspect that as more photos get tagged with general tags (like “landscape”, “man-made”, and “building”), unique and more specific tags will become more valuable. Right now you can imagine that a system implementing ALIPR will have very high recall for these basic words.
For fun I uploaded this photo. These were suggested tags: landscape, man-made, building, train, historical, rural, car, grass, garden, sculpture, estate, people, water, beach, sky.