Ask HN: How could this forgery detection script be optimized?

So, I was reading about this doctored photo in the NYT Magazine:

http://minnesota.publicradio.org/collections/special/columns/news_cut/archive/2009/07/fakery_in_the_times.shtml

And I recalled seeing this blog post and python script a while back:

http://code.google.com/p/coding-experiments/source/browse/trunk/Python/detect_copymove.py

I grabbed the script and image in question from the NYT. I ran the script, and sure enough, several blocks of pixels came back as being copied.

Problem was–the script took almost 5 minutes to analyze the image.

Just curious if it could be optimized or, better yet, easily rewritten in C to be much faster.