How to cluster a large set of object images according to appearance?

edit

Hi,

I have a large collection of flower images. I would like to cluster them to classes according to appearance (not necessarily according to species).

For example, I would like to have a class of blue flowers in a straight pose that have a narrow shape and a different class of red flowers that are in slanted pose and have a more rounded shape.

I'd like that the clustering will be done automatically of course.

I thought about applying k-means to intensity values, perhaps in the HSV color space, but I don't think that will get good results.

Perhaps fitting the images somehow to a Multi-Mixture model? Or regression tress can help (I'm really not that familiar with them, I'm just suggesting ideas).

Can anyone please suggest a method of doing so?

Any ideas will be greatly appreciated!

Thanks,

Gil.