This interesting image recognizer is being developed by James Wang at the College of Information Sciences and Technology at the University of Pennsylvania. You can either upload your own image, or give it an image URL. It analyzes the content of the file, and not the file name.

It seemed to work okay at the demo. However, thinking about it, folks were often grabbing images by doing random google image searches and then plugging in the URL of the image - which maybe tended to bias the demo searches towards more obviously recognizable. I tried it with some of my own photos and it sometimes did less well.

The site also has some image tagging features (and also some of the descriptions supplied by users are tag-like). I’m wondering how far actual machine-based image recognition would be supported or hindered by tags (as tags can be our interpretations/responses to images, as well as descriptions of the actual content of an image).

You can test it out (and help train it) by submitting image URLs at alipr.com.