Consider using absolute instead of relative pixel difference count

[simoncozens]

We currently rank (and filter) the diffs by the number of changed pixels divided by the number of total pixels. The disadvantage of this is that long words can hide significant differences. I'm getting good results just using the absolute count.