using md5sum to avoid CP

Say every time some dude post CP, in addition to permaban the punk, you make a md5sum of the pic (or of several parts of the pic - wiser I think) and store the md5 hash in a database table checked for each pic posted on a board. Obviously even a tiny modification of the pic would make the script useless. Then it could be better to analyze different parts of it. Since the same pics are posted over and over, It could be useful as an autoban feature and pretty easy to write. Obviously it will fail if the pic is cropped or resized.

In the end you'll have 1543 entries for the same reencoded picture and you can't detect duplicates if the poster has half a brain.

everyone knows that if an image isn't accepted, you just fire up an image editor and change a pixel... they already compute hashes to check for duplicates anyway.

>>1
1. What is your problem with CP?
2. As you might have noticed if you were paying attention: no one posts CP pictures on this board. Maybe you should go back to the places which do have CP problem and discuss it there?

>>3
why not calculate the md5 hash of the cropped image instead of the entire image? The hash would change only if the part which wasn't cut is modified.

>>4
Going back to /b/ ? well, no thank.

>>5
then all they need to do is change 1 pixel in that cropped section.


your system will never work, back to /b/.

>>6
      'back to /b/.'

nah I'd rather just shutdown my computer.

>>7
Or do that.

And scaling or reencoding would make this useless.

in this thread: discussion of the porn van as being drafted in the british isles.
\

Do you think tineye computes pictures' md5 ?
If your answer is "no" : how did you have such a bright idea as using md5 ?
Think again : the only thing you (think you) know about is not the only solution.

How would the people posting CP come to the conclusion that you have to change a pixel in the pic if they haven't uploaded it yet, and gotten banned.

>>11
because evading bans are so hard. Do you even have a clue what you're talking about?

Back to /g/, maybe they can teach you some stuff.

I thought about md5 because it is simple. And quite frankly I didn't think more than a minute before creating this thread. But actually it could work if the whole image is not modified (by scaling, luminosity or color modifications,...). You could randomly chose a part of the image from where you compute the hash and store the hash number alongside the 'coordinates' of this part. It prevents to easily spot where to modify the image in order to bypass the filter. But it is a very inelegant solution.

here some ideas: http://stackoverflow.com/questions/596262/image-fingerprint-to-compare-similarity-of-many-images

Treat the image as a matrix and compute its eigenvalues. Use that as a hash.

Common message digestion is the worst way to think about it because it is highly sensitive to entropy in the data used, not the meat of the content. To make this worse, simply resaving a jpeg without any changes will likely produce a different sum due to recompression... but the image will look the same. The thing to focus on is the visual content, not the underlying binary representation. And the kind of digest should not worry so much about the low-order information in the image.

Maybe we should have a /prog/ contest for detecting image similarities. For once we could score them using actual trials, and there's more room for creativity in the solution than our traditional contests, which usually boil down to coding style vs language of choice vs whether or not you are Xarnusing Allegro.

Just take the pixel in the middle of image and ban that, it's very unlikely that two pictures have the same pixel in the middle.

Get Back to /FAGGOT/, Stackoverflow

>>17
Fuck off, ``faggot''.

>>16
That was a good idea (modified to check for a couple of pixels) until modern human has developed the ability to resize and crop images.

>>19
That wouldn't work either. There's no promises that any given pixel will retain the exact same value. >>16 is assuming that the value of the pixel will not normally change significantly. But differences in compression would make the range of values overlap with other images too frequently for a useful test.

ITT PEOPLE WHO THINK BANS ACTUALLY WORK

back to /b/ with you all.

>>21
I'm sure you have some alternatives in mind...

>>22
make a /cp/ board and dedicate it to cp
give FBI direct access to logs

mine away

>>21
Bans are great for culling proxy nodes.

http://www.gnu.org/software/gift/

If you don't know what you're talking about, maybe you should just shut up.

Keyword search http://dis.4chan.org/prog/subject.txt for image matching.

facial and body feature recognition with age determination.
-> sell product

ENTERPRISE PATERNALISM

It would be pretty trivial to write a program that adds a few random characters to the non-image section any jpeg fed into it. You don't have to change pixels.Just use the concept behind imgraring.

I wrote a script that used that for image spamming. It was really simple.

I think this is all a ploy for OP to get CP in the claims that he'll create a checksum of 'chunks' to prevent people from posting the CP images he has ahold of