Image Recognition using Supervised or Unsupervised Learning

I am familiar with supervised Learning methods (SVM, Maximum Entropy, Bayes Classifiers) for textual classification, but for image I cannot figure out where I should start from.

I have a set of human images (exclusively women) whom I've to classify as being beautiful or not. The first hurdle I am facing is "Feature selection". I though开发者_Go百科t to take hair shape, complexion, eye shape as features but they are becoming too complex to detect. OCR in comparison seems comparatively easier as the shapes can be put in black & white format and find best match with the known symbols.

I am also ready to explore unsupervised learning methods if that is more useful. Please provide me pointers as to how should I begin with. Any free to use libraries would be really great (could be in any language)!

If you want to classify beatiful/not beautiful, you want to use supervised learning because unsupervised learning has no idea about the concept "beautiful", you have to introduce that via labels.

As a pointer to get you started, there is a recent paper on facial beauty on ECCV 2010 (you can find it on http://dgray.info/), in the introduction there are lots of references to other papers.