I am trying a variety of things right now... Small size and oddly shaped covers are pretty easy to find. Finding covers that are mostly white is pretty easy but slow. What I'm doing now is down sampling the cover and then building clusters of color. Then I look at the clusters and determine the prevalent color. The problem with that is that it does generate a lot of false positives. In the past I did some work with Bayesian probabilities and I think this might be a way to determine good versus bad covers.
I'm not so sure about the focus thing yet. I'm still exploring.
Hopefully I'll have something to show in the next week or two.