scikit-image: segmentation and regionprops

We show how to segment a photo of coins, separating the foreground from the background. Our process is to denoise the image (using a median filter), and to then apply watershed segmentation. The resulting segments are cleaned up, using region properties and K-means clustering. There are certainly many other ways of approaching this problem, e.g. region agglomeration with region adjacency graphs, supervised clustering, etc. Please share your ideas in the comments below!