CP detection

[Yes, I know it means Politically Correct, but how much does it cost to type a few more letters and make the post more easily readable?

I'll try to avoid acronyms this time.

Writing out both acronyms would have lost the chiastic structure of the description, which I thought was more important to retain in this particular case.

Besides, I felt reasonably sure you already had found this

http//www.forensicfocus.com/Forums/viewtopic/p=6522400/#6522400

I'll try to avoid acronyms this time.

Writing out both acronyms would have lost the chiastic structure of the description, which I thought was more important to retain in this particular case.

Besides, I felt reasonably sure you already had found this

http//www.forensicfocus.com/Forums/viewtopic/p=6522400/#6522400

Yep, and sorry for being a nuisance oops , but I think the board posts should be as clear as possible ) .
In theory 😯

A PC using a PC to inspect another PC owned by a CP to find CP, possibly inside a PC, and wearing PC while inside a PC building - authorized by a CP along a plan developed by a CP -, could send a PC ( with the proper PC and address in PC) written in PC terms to the local PC (which does have a few members from CP) which could call for a PC, people attending will get there either in a PC or in a PC, the CP being the speaker. Surely the CP reporters will be present.

1. PC=Private Contractor
2. PC=Personal Computer
3. PC=Personal Computer
4. CP=Cyber Punk
5. CP=IIOC wink or CEP
6. PC=Private Chat
7. PC=Protective Clothes
8. PC=Precast Concrete
9. CP=Construction Permit
10. CP=City Planner
11. PC=Post Card
12. PC=Postal Code
13. PC=Panama City
14. PC=Politically Correct
15. PC=Political Committee
16. CP=Communist Party
17. PC=Press Conference
18. PC=Patrol Car
19. PC=Passenger Car
20. CP=Commissioner of Police
21. CP=Cincinnati Post

[/listo]

jaclaz

That post is a work of art.

*prints and hangs on office wall*

*prints and hangs on office wall*

JFYI, that post taken out of context might represent a SERIOUS offence under provisons of section 1.1 😯
http//www.acronympolice.org/
http//www.acronympolice.org/?page_id=90
(you should put a note below it stating how it is for satirycal purpose, to be on the safe side. wink )

jaclaz

Even Department of Defense employees are subject to the laws in their jurisdiction. If child pornography is located and you are not a sworn law enforcement agent acting in the official capacities of ones duties then it is illegal for you to view it (in my jurisdiction anyway and under federal law). If CP is located, I would suggest immediately stop viewing it, remove the device and then report this activity to a supervisor as well as seek law enforcement advice/open a joint case. For more reasons than just one as I highly doubt the DOD is going to prosecute the indivdual responsible for putting it there in the first place.

James Poole, CFCE

If CP is located, I would suggest immediately stop viewing it, remove the device and then report this activity to a supervisor as well as seek law enforcement advice/open a joint case.

Law enforcement has been notified and have taken over the investigation

roll

jaclaz

Trying to steer back to the original question…

I know of skin tone detection, but body parts?

How would you detect body parts? Edge detect, then try to match the shape to a library of shapes?

How would you detect body parts? Edge detect, then try to match the shape to a library of shapes?

Yes, that would be theory, but how many shapes (or positions) can a human body assume?
I presume that if you first go for the skin tone, then you use something like (cannot say if it exists) "filter" for "amount of skin" in the picture (just to quickly rule out "fully dressed" photos and portraits) you won't have that many photo's remaining.

There will be still an issue with wink
http//vulcanstev.files.wordpress.com/2011/11/demot-green.jpg
but I guess it's a rare case.

More seriously, what about level of accurateness proportionated to numbers?
I mean, let's say that you have two cases, case A) and Case B).
Case A)
You find with your triage/autodetecting tool (say) 217 "suspect" images, which you go through "manually" and once eliminated the false positives (say) 59, you have 158 incriminating images.
BUT, on further examination (one by one of the total - still say - 1285 images) you find another 12 incriminating images.
The tool(s) have definitely a very high accuracy, as it identified among the 1285 images 217 and of these 158 are actually relevant. 158/217=72.8% and 158/170=92.9%
I presume that in this case if the suspect is incriminated (and later condemned) for having 158 such images instead of for having 170 makes no difference.

Case B)
The tool finds 134 "suspect images" (out of a total of 2784), when you go manually through these 134 you find that noone represents CP or IIOC.
What do you do next?
Maybe if you go manually through the images, you can find 8 of the same images as in case A) above (at different resolutions/with some marginal changes that modifies the hashes, so that you cannot use your database of hashes).
There is the risk of having the suspect cleared for what is otherwise a "marginal" error in the working of the software.

How accurate is the "manual" process normally (I presume that there is anyway an area of possible mistakes - human generated)?

The tool should guarantee a level of accuracy similar to the "manual" way, net of the possible human errors, but how much can they be?

I presume that the ones more interested to "identify human shapes" are the Military, for automated/computerized weapons/defense systems and for security/video surveillance, but there is seemingly a lot of interest (just like for facial recognition)
http//www.umiacs.umd.edu/~wamageed/papers/icra2008.pdf
http//sdpy.googlecode.com/svn/tags/temp/unsorted3/to_delete/tmp/unsorted/to_remove/research/my_papers/tag/phdthesis_1st_submission_in_May/review/survey/component-based/138_iccv07.pdf
http//citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.154.7771&rep=rep1&type=pdf
http//www.cs.brown.edu/~black/Papers/freifeldCVPR10.pdf
http//www.ri.cmu.edu/pub_files/pub4/zhao_liang_2001_1/zhao_liang_2001_1.pdf

There is a whole (wider) field of study called CBIR
http//en.wikipedia.org/wiki/Content-based_image_retrieval
It is very likely that some of the projects will (or may) evolve in the direction of recognizing "poses"
http//en.wikipedia.org/wiki/List_of_CBIR_engines

jaclaz

percent skin-tone is not hard, and is already available.
We can write a script for GIMP to do it…
The problem with this is, that not always indicative of negative outcome.
i.e. low % of skin-tone in image is not necessarily not inappropriate material.

I have seen algos for human shape detection, but only in motion. I have seen partial image matching, but the original image was not squishy (as in body), but fixed (think sub or ship).

The best matching I have seen was in food processing industry. High speed cameras looking at fruit as they fly by. Code recognizes bruised, and not properly shaped…

Code recognizes bruised, and not properly shaped…

Oww, come on, how many "poses" can (say) a pear have? roll

jaclaz