Getting different hashes on bit-for-bit identical PNGs

[McGeekiest]

With two files, shown to be identical via diff, I am getting dramatically different hash values:

$ diff 0_0.png 0_1.png 
$ ruby

require 'rubygems'
require 'phashion'
p1 = Phashion::Image.new('0_0.png')
p2 = Phashion::Image.new('0_1.png')
puts p1.duplicate?(p1) ? "true" : "false"
puts p1.duplicate?(p2) ? "true" : "false"
puts "p1 fingerprint: " + p1.fingerprint.to_s
puts "p2 fingerprint: " + p2.fingerprint.to_s
puts (p1.fingerprint ^ p2.fingerprint).to_s(2).count("1")

Results in:

true
false
p1 fingerprint: 6242629367104429399
p2 fingerprint: 957486519169142254
38

Am I doing something wrong?

[westonplatter]

@McGeekiest which phashion version are you on?

[McGeekiest]

@westonplatter 1.0.6

[westonplatter]

ah, I think I know what the problem is.

We made a custom patch to the pHash C library that was released in gem version 1.0.8. The README describes this more, https://github.com/westonplatter/phashion#gem-uses-customized-phash-096.

I'll be pushing an updated 1.1 version this weekend. Until then, you could reference the GitHub master branch. In case it's helpful, more info about how to do that here, #32 (comment).

[McGeekiest]

Thanks for the help so far, @westonplatter!

$ bundle show | grep phashion
  * phashion (1.0.8 2c6d8c9)

... then I get the same problem in a Rails console. Curiously, the fingerprints on both files differ every time I run the code, as do the hamming distances. I thought maybe there was just an invocation problem; glad to know I wasn't missing something obviously obvious (as it were). :-)

Can I provide any other information to help? I'm on Rails 4.0.2 under Ubuntu Saucy, to start with.

[McGeekiest]

Oh, hello! Based on the hints in the README, I tried:

$ convert 0_0.png -alpha remove 1_0.png
$ convert 0_1.png -alpha remove 1_1.png

Now it works on the two new PNGs. It's apparently an alpha thing. I'll just write a script to strip alpha information from the PNGs.

[westonplatter]

Nice solution.

Sorry @McGeekiest. That's frustrating. I'll have test that and add tests to prove we're catching the alpha-png/regular pngs. Could you send the png images you're using to test? email = my github username [at] gmail

[McGeekiest]

@westonplatter Done. Let me know if I can help further.

[westonplatter]

@McGeekiest tested alpha png images you sent locally, and Phashion said they were duplicates. :(.

I'll update the readme, and talk with @toy to see if he's had coontinued problems with alpha PNGs.

Thanks for the help!

[toaco]

I'am using phashion 1.2.0 and have encountered a similar problem, the same png image has a different fingerprint, and the console occasionally output:

libpng warning: iCCP: known incorrect sRGB profile