[jemhoff]

How I'd do it: Use Embedly! Our extract API returns the dominant colors of all the images it finds: http://embed.ly/docs/extract/api/response#extract-images

For those that are curious, what happens under the covers is a random sampling of the image followed by a conversion to LAB color space (even better than YUV in terms of human perception) and finally some K-means clustering.

To see this in action, check out http://tightro.pe.