Per-Title Encode Optimization

I'd be curious to read about the software that does the PSNR analysis, or if anyone here has done some automated experiments of their own. I'm constantly looking to improve encoding of screencast-style video, which can be tricky to make assumptions about due to source material that sometimes has windows full of code being scrolled quickly.

Given a source file and, say, a series of ffmpeg encodes at various resolutions/bitrates, is there a simple(ish) way of automated perceptual quality scoring, a'la PSNR or another objective "proxy" variable that closely correlates with the subjective perception of video quality? Maybe representative thumbnails taken from every n-th frame and compared?

I guess that last point would expand the possibilities to include automated static image comparisons, since the script to get ffmpeg (or Elastic Transcoder, for that matter) to save out thumbnails is trivial.

Having all of the images gated behind a Google Docs access request: accident or attempted recruitment lead generator?