I remember this issue being brought up before, specifically with account numbers.  As it turns out, with blurred images with a finite set of possibilities (i.e. obscuring numbers rather than an image), its not too awful hard to determine the obfuscated numbers.  This link describes a method for doing just that, specifically for images delivered digitally.

http://dheera.net/projects/blur.php


--
Brian Adams
Technical Director
33 Auditorium
Department of Theatre
Michigan State University
East Lansing, MI 48824


On 10/23/07, Richard Wiggins < [log in to unmask]> wrote:
Recently police in Thailand arrested an accused pedophile after German detectives reconstructed an image of the suspect's face that had been distorted using image editing software.
 
Clever work on the part of the German detectives.  This got me to thinking about managing sensitive data implications.  If you distribute an image, part of which you've blacked out -- say, a name or an SSN -- unless the transformation is truly one-way, someone could do the same thing.
 
This could be in a PowerPoint presentation or any document.
 
I'm thinking the only really safe way to do this would be to blacken a printed copy (or better yet cut out the sensitive info) and scan it back in.
 
Food for thought...
 
http://www.smh.com.au/news/world/trace-on-boyfriends-mobile-leads-police-to-pedophile-suspect/2007/10/19/1192301044777.html
 
/rich