Buckinghamshire CC ANPR cameras
tugwilson at gmail.com
Mon Jan 9 16:34:51 GMT 2012
On 9 January 2012 15:26, Clive D.W. Feather <clive at davros.org> wrote:
> Chris Edwards said:
>> I guess hashing a registration is pretty easy to reverse. With access to
>> the DVLA database of all UK registrations, you simply compute the hash for
>> each one. This is a one off task. Once you've done it, you know the
>> registration for any given hash.
I've turned up some more information on the hashing mechanism.
The Highways Agency say "The processor identifies the number plate in
each image and uses optical character recognition to obtain the plate.
The plate is then simplified and hashed with a large prime number, to
give a non-reversible tag."
This paper from Sheffield University says "a process known as
“character merging” is applied prior to generation of the hash value
from the plate characters. In this process, character sets which are
commonly mis-read by ANPR systems (such as O, D and 0, or 8, B and 3)
are merged (e.g. all Ds are replaced by Os etc) which has the effect
of reducing significantly the uniqueness of the hash value"
However the number plate layout introduced in September 2001 should
not require as much merging as the letters and digits are readily
distinguishably by position.
The mention of the prime number leads me to guess they are using the
sum of products method to create the hash.
As the hash is only 24 bits and there are 34.5 million licensed
vehicles on the road there must be collisions even if the hash
function were perfect (and it's not).
It would be fun to run all the currently issued licence numbers
through the hash function to see how well it works. Unfortunately the
DVLA won't supply a list of the current licence plate numbers and the
Highways Agency won't disclose the hashing algorithm.
It seems clear that this method as a very poor way of anonymising the
data. If I had access to the data I could easily identify and
individual's car with out knowing anything at all about the hash
function other than it always produces the same hash given the same
number plate. I just need to know the precise time they pass one or
more cameras or less precise information on several journeys.
More information about the ukcrypto