You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Luis Daniel Lucio Quiroz <lu...@gmail.com> on 2009/03/02 00:19:47 UTC
Hopfield nerons for porn image detection
HI all,
While studing a subject of hopfield for my master. Reading, I got that a
Hopfield network is able to detect patterns even if inputs have noise. So for
example many people could write A capital in many ways and hopfield network
will detect A symbol.
What do you think on using this to detect porn images for incoming email. I
mean, after a long good email training this could be possible.
Any comment?
LD
Re: Hopfield nerons for porn image detection
Posted by Karsten Bräckelmann <gu...@rudersport.de>.
On Tue, 2009-03-03 at 16:36 -0600, Luis Daniel Lucio Quiroz wrote:
> Yes,
>
> but linked is detected over other solutions related to websurfing.
>
> I'm focusing into mail attachment, those nothing-better-to-do users that sends
> images to his friends using their corporate email.
Whoops, wrong list, eh? :) Are you talking about the Powerpoint presen-
tations containing low-res images, badly animated? Yes, they are
annoying, but no spam.
> Even, this could be combined with other modules that let SA download image and
> then analize them.
Which, as we explained, can and *is* being exploited to validate
addresses to send even more spam. Or sell them. Hey, verified addresses
are more valuable than made-up ones...
Following links automatically is quite a bad and potentially harmful
idea.
> On Tuesday 03 March 2009 15:14:51 RW wrote:
> > But as Karsten said, porn images are normally linked, rather than
> > attached, and links shouldn't be followed in case they are matched
> > to the email address.
Top-posting retained...
--
char *t="\10pse\0r\0dtu\0.@ghno\x4e\xc8\x79\xf4\xab\x51\x8a\x10\xf4\xf4\xc4";
main(){ char h,m=h=*t++,*x=t+2*h,c,i,l=*x,s=0; for (i=0;i<l;i++){ i%8? c<<=1:
(c=*++x); c&128 && (s+=h); if (!(h>>=1)||!t[s+h]){ putchar(t[s]);h=m;s=0; }}}
Re: Hopfield nerons for porn image detection
Posted by Karsten Bräckelmann <gu...@rudersport.de>.
On Tue, 2009-03-03 at 15:16 -0800, John Hardin wrote:
> On Tue, 3 Mar 2009, Luis Daniel Lucio Quiroz wrote:
>
> > I'm focusing into mail attachment, those nothing-better-to-do users that
> > sends images to his friends using their corporate email.
>
> Ah, corporate policy enforcement.
>
> I wouldn't worry so much about trying to intelligently identify images to
> filter out porn. If you have a user who is abusing the system that way
> they should be easy to identify through other means.
Yeah, agreed. Moreover, SA typically is best run on the receiving end.
The abusive user is the sender...
--
char *t="\10pse\0r\0dtu\0.@ghno\x4e\xc8\x79\xf4\xab\x51\x8a\x10\xf4\xf4\xc4";
main(){ char h,m=h=*t++,*x=t+2*h,c,i,l=*x,s=0; for (i=0;i<l;i++){ i%8? c<<=1:
(c=*++x); c&128 && (s+=h); if (!(h>>=1)||!t[s+h]){ putchar(t[s]);h=m;s=0; }}}
Re: Hopfield nerons for porn image detection
Posted by John Hardin <jh...@impsec.org>.
On Tue, 3 Mar 2009, Luis Daniel Lucio Quiroz wrote:
> I'm focusing into mail attachment, those nothing-better-to-do users that
> sends images to his friends using their corporate email.
Ah, corporate policy enforcement.
I wouldn't worry so much about trying to intelligently identify images to
filter out porn. If you have a user who is abusing the system that way
they should be easy to identify through other means.
A couple of possibilities: collect statistics on how much image traffic
(bytes) each user sends, and how much image data is stored on their
workstation. With that data you'll be able to focus on the problem users.
Not sure that SA is the correct tool for this.
--
John Hardin KA7OHZ http://www.impsec.org/~jhardin/
jhardin@impsec.org FALaholic #11174 pgpk -a jhardin@impsec.org
key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
Failure to plan ahead on someone else's part does not constitute
an emergency on my part. -- David W. Barts in a.s.r
-----------------------------------------------------------------------
5 days until Daylight Saving Time begins in U.S. - Spring Forward
Re: Hopfield nerons for porn image detection
Posted by Luis Daniel Lucio Quiroz <lu...@gmail.com>.
Yes,
but linked is detected over other solutions related to websurfing.
I'm focusing into mail attachment, those nothing-better-to-do users that sends
images to his friends using their corporate email.
Even, this could be combined with other modules that let SA download image and
then analize them.
On Tuesday 03 March 2009 15:14:51 RW wrote:
> On Tue, 3 Mar 2009 14:19:24 -0600
>
> Luis Daniel Lucio Quiroz <lu...@gmail.com> wrote:
> > Hopfield networks are not the fastest, but they can identify paterns
> > with noise. I was wondering to cut an image in smaller zones nxn and
> > then run hopfield network to detect naked body parts. After
> > detectin these, a grade could be done.
>
> But as Karsten said, porn images are normally linked, rather than
> attached, and links shouldn't be followed in case they are matched
> to the email address.
Re: Hopfield nerons for porn image detection
Posted by RW <rw...@googlemail.com>.
On Tue, 3 Mar 2009 14:19:24 -0600
Luis Daniel Lucio Quiroz <lu...@gmail.com> wrote:
>
> Hopfield networks are not the fastest, but they can identify paterns
> with noise. I was wondering to cut an image in smaller zones nxn and
> then run hopfield network to detect naked body parts. After
> detectin these, a grade could be done.
But as Karsten said, porn images are normally linked, rather than
attached, and links shouldn't be followed in case they are matched
to the email address.
RE: Hopfield nerons for porn image detection
Posted by Michael Hutchinson <mh...@manux.co.nz>.
-----Original Message-----
From: Luis Daniel Lucio Quiroz [mailto:luis.daniel.lucio@gmail.com]
Sent: Wednesday, 4 March 2009 9:19 a.m.
To: users@spamassassin.apache.org
Subject: Re: Hopfield nerons for porn image detection
Good one,
Hopfield networks are not the fastest, but they can identify paterns
with
noise. I was wondering to cut an image in smaller zones nxn and then
run
hopfield network to detect naked body parts. After detectin these, a
grade
could be done.
Any comment.
------------------------------------------------------------
Yes - Why even bother detecting naked body parts when there is little
point? Others on this list have already commented about the fact that
there is very little image Spam out there nowdays - in fact, is there
any?, and what there is uses external links to display it's content,
there are no images in the actual E-Mail message, or attached to it.
I guess if you really wanted naked body part image matching work, you
could write another Net-Nanny style program to filter websites for
families...
2c
Cheers,
Mike
Re: Hopfield nerons for porn image detection
Posted by Luis Daniel Lucio Quiroz <lu...@gmail.com>.
Good one,
Hopfield networks are not the fastest, but they can identify paterns with
noise. I was wondering to cut an image in smaller zones nxn and then run
hopfield network to detect naked body parts. After detectin these, a grade
could be done.
Any comment.
On Sunday 01 March 2009 18:49:16 Kai Schaetzl wrote:
> Karsten Bräckelmann wrote on Mon, 02 Mar 2009 00:35:35 +0100:
> > Please excuse my ignorance -- how fast is that? Also, there's quite a
> > difference between identifying an "A" symbol and porn, isn't it?
>
> Depends on how well the bodies form the alphabet ;-)
>
> Kai
Re: Hopfield nerons for porn image detection
Posted by Kai Schaetzl <ma...@conactive.com>.
Karsten Bräckelmann wrote on Mon, 02 Mar 2009 00:35:35 +0100:
> Please excuse my ignorance -- how fast is that? Also, there's quite a
> difference between identifying an "A" symbol and porn, isn't it?
Depends on how well the bodies form the alphabet ;-)
Kai
--
Kai Schätzl, Berlin, Germany
Get your web at Conactive Internet Services: http://www.conactive.com
Re: Hopfield nerons for porn image detection
Posted by Karsten Bräckelmann <gu...@rudersport.de>.
On Sun, 2009-03-01 at 17:19 -0600, Luis Daniel Lucio Quiroz wrote:
> HI all,
>
> While studing a subject of hopfield for my master. Reading, I got that a
> Hopfield network is able to detect patterns even if inputs have noise. So for
> example many people could write A capital in many ways and hopfield network
> will detect A symbol.
>
> What do you think on using this to detect porn images for incoming email. I
> mean, after a long good email training this could be possible.
Please excuse my ignorance -- how fast is that? Also, there's quite a
difference between identifying an "A" symbol and porn, isn't it?
Anyway, I guess an important show-stopper here is, that image spam is
seriously low these days. That is, a message containing the payload of
an image. Sure, there is much more spam containing links to images --
but fetching them for investigation is potentially harmful, since that
stuff can be and is easily abused to verify live addresses.
Just some quick thoughts...
guenther
--
char *t="\10pse\0r\0dtu\0.@ghno\x4e\xc8\x79\xf4\xab\x51\x8a\x10\xf4\xf4\xc4";
main(){ char h,m=h=*t++,*x=t+2*h,c,i,l=*x,s=0; for (i=0;i<l;i++){ i%8? c<<=1:
(c=*++x); c&128 && (s+=h); if (!(h>>=1)||!t[s+h]){ putchar(t[s]);h=m;s=0; }}}