>>175
Feels pretty tedious. If you calculate the hamming distance between the two hashes, you'll see that's 2. If jschan could do it, you could just add a filter for ff00c1d57a61a8ba and (say) max distance=2, and it would match both. And also ff00c1d57a61acbe, which is also bound to be similar.
My "hashes" look like this:
	'CP1': [
		0.489656, 0.0675636,  1.60258,	 1.42863,
		3.52474,   1.38529,	 3.44002,	1.75575,
		6.98833,	3.4452,	  4.3161,	2.47032,
		7.21341,   3.51376, 0.424894, 0.0113652,
		1.53821,  0.507692,	 3.73382,	1.27282,
		3.30786,   1.60575,	 6.93337,	3.04631,
		4.18564,   1.87194,	  7.0374,	4.16202,
		0.352069,  0.437672,  1.49136,	 1.54866,
		4.65797,   3.70002,	 3.47888,	3.36337,
		7.54734,   6.99339,	 4.76161,	4.24589,
		9.50999,	7.1144
	],
Which is a pretty unholy sight, but this is what imagemagick produces. Then I can calculate the mean squared error and block if it is too high. But the current solution is pretty slow and resource heavy, so I will likely explore other options in the future.