I basically developed it because of Wii Tennis. Those thin white lines are very challenging and I was noticing artifacts with the usually excellent "Motion Adaptive: Advanced Detection" algorithm. I tried a few things and what I came up with is a pretty simple concept: look for combing artifacts. It only requires two fields.
Basically it examines 5 vertical pixels (I call them a,b,c,d,e) with the middle pixel, c, the one being deinterlaced. If c is between b and d then weave (output c). If b is between a and c, and d is between c and e then weave (i.e. c is the top of a hill or bottom of a valley). Otherwise deinterlace and output the average of b and d.
This gives an very good sharp picture for static images and no artifacts for fast moving images.
You'll have to run autogen.sh after applying this patch, as it adds a new source file to the makefiles.