I have now, thanks to you, and I think I
probably owe you a beer.
I ran the visual similarity search on about
4,700 images. The results were overly
generous, but there are about half a dozen
algorithm tweaks that I haven't tried playing
with yet.
It did a good job finding cropped, resized,
and color-adjusted similarities; it did a bad
job identifying images with captions or logos
added/removed.
I also ran the exact duplicate search on
17,000+ images, and the program, just
like Tiny Tim on the last page of A Christmas
Carol, did NOT die.
A tip of the hat to charliex!