Testing visual regressions is pretty hard. Here is an introduction to PhantomCSS, aimed to solve this problem.
See in Fullscreen