Sign in with
Sign up | Sign in
Your question

Zoom in and make clearer like they do on TV

Last response: in Digital Camera
Share
Anonymous
March 27, 2005 11:06:02 PM

Archived from groups: rec.photo.digital (More info?)

Watching TV shows such as CSI they will Zoom in and make picture clear
enough to read licensev plate numbers how do they do that?

Thanks
Mark



----== Posted via Newsfeeds.Com - Unlimited-Uncensored-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----

More about : zoom make clearer

Anonymous
March 27, 2005 11:06:03 PM

Archived from groups: rec.photo.digital (More info?)

Mtco wrote:
> Watching TV shows such as CSI they will Zoom in and make picture
> clear enough to read licensev plate numbers how do they do that?
>

Short answer: They don't. It's Science Fiction. Beam me outta there.

I usually have great patience with repeating threads that teach
something. Not this one. It always turns into a "Yeah, remember when
Fulano such-and-suched! Do they think we're idiots?" contest.
Interminably.

If that's what you enjoy, search Google groups on that theme. Rich vein
awaiting you.

--
Frank ess
March 27, 2005 11:06:03 PM

Archived from groups: rec.photo.digital (More info?)

Mtco wrote:

> Watching TV shows such as CSI they will Zoom in and make picture clear
> enough to read licensev plate numbers how do they do that?
>
> Thanks
> Mark


I think there are some lenses with constant zoom so you can zoom in to
focus then zoom back to re-frame but it's pretty unusual.
Related resources
Anonymous
March 27, 2005 11:06:03 PM

Archived from groups: rec.photo.digital (More info?)

Bottom line is it's fake. The digital information just is not there to
zoom in to get a clear picture of something that is grainy. Some
programs may be able to guess/approximate/interpolate the missing data,
but it's just a best estimate, and probably not to the degree they lead
you to believe.
Anonymous
March 27, 2005 11:41:25 PM

Archived from groups: rec.photo.digital (More info?)

Are you saying it can't be done?
"Frank ess" <frank@fshe2fs.com> wrote in message
news:ypednU8aA7pSwNrfRVn-ig@giganews.com...
> Mtco wrote:
>> Watching TV shows such as CSI they will Zoom in and make picture
>> clear enough to read licensev plate numbers how do they do that?
>>
>
> Short answer: They don't. It's Science Fiction. Beam me outta there.
>
> I usually have great patience with repeating threads that teach something.
> Not this one. It always turns into a "Yeah, remember when Fulano
> such-and-suched! Do they think we're idiots?" contest. Interminably.
>
> If that's what you enjoy, search Google groups on that theme. Rich vein
> awaiting you.
>
> --
> Frank ess
>
>
>



----== Posted via Newsfeeds.Com - Unlimited-Uncensored-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----
Anonymous
March 28, 2005 12:28:32 AM

Archived from groups: rec.photo.digital (More info?)

Mtco wrote:
> Watching TV shows such as CSI they will Zoom in and make picture clear
> enough to read licensev plate numbers how do they do that?
>
> Thanks
> Mark
>
>
>
> ----== Posted via Newsfeeds.Com - Unlimited-Uncensored-Secure Usenet News==----
> http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
> ----= East and West-Coast Server Farms - Total Privacy via Encryption =----

They do that with 'trick photography'. There is some rather
sophisticated software used by the NSA, but you couldn't afford the
computers it runs on. Beyond that, forget it.
It's just for TV/Movies.


--
Ron Hunter rphunter@charter.net
Anonymous
March 28, 2005 12:44:59 AM

Archived from groups: rec.photo.digital (More info?)

In the old days - lots of expensive computer capacity and well
designed algorithms. Today - a fast computer processor, lots of
memory, and a well designed algorithm.
Probably first conducted by military /NSA-types for surveillance
photos from aircraft /space - the process was significantly enhanced
by the propeller-heads at places like Jet Propulsion Labs to produce
those great space photos from space probes to Hubble to the Mars
voyagers. Pretty amazing stuff - can be tailored to add /subtract
/enhance / artifically color, etc. all or specific spectra / types of
info.
As for reading license plates - as you 'zoom' in blanks in information
(image) become (painfully) evident. A well constructed algorithm could
'best fit' probable fractals into the voids to artifically create
information re: enhance resolution.
An interesting product, available to consumers (and not all that
expensive for what it does) is called GenuineFractals
(genuinefractals.com) It allows enlargement of even small photo files
up to 700% (according to the web site) without loss of resolution. A
more enhanced version it available for photo professionals. Haven't
tried it - but it sounds interesting.



>enough to read licensev plate numbers how do they do that?
>
>Thanks
>Mark
>
>
>
>----== Posted via Newsfeeds.Com - Unlimited-Uncensored-Secure Usenet News==----
>http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
>----= East and West-Coast Server Farms - Total Privacy via Encryption =----
Anonymous
March 28, 2005 12:45:00 AM

Archived from groups: rec.photo.digital (More info?)

cbrs@silicondairy.net wrote:
> In the old days - lots of expensive computer capacity and well
> designed algorithms. Today - a fast computer processor, lots of
> memory, and a well designed algorithm.
> Probably first conducted by military /NSA-types for surveillance
> photos from aircraft /space - the process was significantly enhanced
> by the propeller-heads at places like Jet Propulsion Labs to produce
> those great space photos from space probes to Hubble to the Mars
> voyagers. Pretty amazing stuff - can be tailored to add /subtract
> /enhance / artifically color, etc. all or specific spectra / types of
> info.
> As for reading license plates - as you 'zoom' in blanks in information
> (image) become (painfully) evident. A well constructed algorithm could
> 'best fit' probable fractals into the voids to artifically create
> information re: enhance resolution.
> An interesting product, available to consumers (and not all that
> expensive for what it does) is called GenuineFractals
> (genuinefractals.com) It allows enlargement of even small photo files
> up to 700% (according to the web site) without loss of resolution. A
> more enhanced version it available for photo professionals. Haven't
> tried it - but it sounds interesting.
>
>
>
>
>>enough to read licensev plate numbers how do they do that?
>>
>>Thanks
>>Mark
>>
>>
>>
>>----== Posted via Newsfeeds.Com - Unlimited-Uncensored-Secure Usenet News==----
>>http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
>>----= East and West-Coast Server Farms - Total Privacy via Encryption =----
>
>
Interesting, but disappointing.


--
Ron Hunter rphunter@charter.net
Anonymous
March 28, 2005 1:12:14 AM

Archived from groups: rec.photo.digital (More info?)

Ah, the old 'hardware' vs. 'software' conundrum...
I think the old Perkin Elmer approach (high tech lenses /
berrylium mirrors) to enhance resolution was great in it's day - but
it's hard to argue against computers /algorithms to 'fill in the
blanks'. Not only can they do that, but they can offer differing
results based on degree of probability desired. In come cases the
results are not all that different - in others, the results can vary
dramatically (hence the need for human evaluators to place the
possibilities in proper perspective re: in consideration of other
evidence). License plates are not all that difficult of a subject.
Uniform size, background color, font, font color. Not all that many
variables (relatively) compared say to a rock on a planet millions of
miles away.

The technology is not 'science fiction.' It's used in countless
fields today - from anthropology to cartography to industrial design
and non-destructive testing (I miss the old days when engineers
actually blew things up to see how well they were built). A good
example is something seen commonly on TV - an engineer or animator
drawing a wire-model, pressing a button - and a smoothed, final
version appears. Just an algorithm filling in spaces (voids) based on
probability (and the criteria entered by the user).


>
>I think there are some lenses with constant zoom so you can zoom in to
>focus then zoom back to re-frame but it's pretty unusual.
Anonymous
March 28, 2005 1:25:21 AM

Archived from groups: rec.photo.digital (More info?)

"Mtco" <news@mtco.com> writes:

> Watching TV shows such as CSI they will Zoom in and make picture clear
> enough to read licensev plate numbers how do they do that?

Ummm, they are telling a story, and don't let hard facts get in the way?

With proper sharpnening and interpolation you get some improvement, but not to
the extent they show on TV.

--
Michael Meissner
email: mrmnews@the-meissners.org
http://www.the-meissners.org
Anonymous
March 28, 2005 4:27:37 AM

Archived from groups: rec.photo.digital (More info?)

cbrs@silicondairy.net wrote:
> Ah, the old 'hardware' vs. 'software' conundrum...
> I think the old Perkin Elmer approach (high tech lenses /
> berrylium mirrors) to enhance resolution was great in it's day - but
> it's hard to argue against computers /algorithms to 'fill in the
> blanks'. Not only can they do that, but they can offer differing
> results based on degree of probability desired. In come cases the
> results are not all that different - in others, the results can vary
> dramatically (hence the need for human evaluators to place the
> possibilities in proper perspective re: in consideration of other
> evidence). License plates are not all that difficult of a subject.
> Uniform size, background color, font, font color. Not all that many
> variables (relatively) compared say to a rock on a planet millions of
> miles away.
>
> The technology is not 'science fiction.' It's used in countless
> fields today - from anthropology to cartography to industrial design
> and non-destructive testing (I miss the old days when engineers
> actually blew things up to see how well they were built). A good
> example is something seen commonly on TV - an engineer or animator
> drawing a wire-model, pressing a button - and a smoothed, final
> version appears. Just an algorithm filling in spaces (voids) based on
> probability (and the criteria entered by the user).

Usually NOT probability, but actual math and application of the surface
design specified.

>
>
>
>>I think there are some lenses with constant zoom so you can zoom in to
>>focus then zoom back to re-frame but it's pretty unusual.
>
>


--
Ron Hunter rphunter@charter.net
Anonymous
March 28, 2005 5:33:40 AM

Archived from groups: rec.photo.digital (More info?)

"Frank ess" <frank@fshe2fs.com> wrote in message
news:ypednU8aA7pSwNrfRVn-ig@giganews.com...
> Mtco wrote:
>> Watching TV shows such as CSI they will Zoom in and make picture
>> clear enough to read licensev plate numbers how do they do that?
>>
>
> Short answer: They don't. It's Science Fiction. Beam me outta there.
>
//snip//

And you know this because?
Anonymous
March 28, 2005 5:33:41 AM

Archived from groups: rec.photo.digital (More info?)

Toomanyputters wrote:
> "Frank ess" <frank@fshe2fs.com> wrote in message
> news:ypednU8aA7pSwNrfRVn-ig@giganews.com...
>
>>Mtco wrote:
>>
>>>Watching TV shows such as CSI they will Zoom in and make picture
>>>clear enough to read licensev plate numbers how do they do that?
>>>
>>
>>Short answer: They don't. It's Science Fiction. Beam me outta there.
>>
>
> //snip//
>
> And you know this because?
>
>
Because some of us keep up with the latest innovations in computers and
digital image processing. To do anything even similar (and most of what
they show is downright impossible) to what you see requires
supercomputers, multiple views from multiple angles, and some software
you would need tons of ram to run. Don't expect to get that running on
your system this week, or even next year. Come back in 10 years, and I
may be able to tell you what you need. For now little can be done short
of governmental budget figures.


--
Ron Hunter rphunter@charter.net
Anonymous
March 28, 2005 5:33:42 AM

Archived from groups: rec.photo.digital (More info?)

.. For now little can be done short
>of governmental budget figures.

Can't agree with that....

According to the latest (Nov 2004) rating of the world's 500 fastest
supercomputers - (see "Top 500 Supercomputer Sites") - #7 is the
'home built' machine at Virginia Tech.
1100 Apple G5 machines as a MPC. An absolute bargain at
$2,060,000.00! ($2 mil for the machines / 60K to put it together).
Bigger bang for the buck than dozens of lesser rated machines at
government labs, nasa, etc. - and at a school that isn't in the same
league as MIT, CalTech, etc. with all their government programs
/support. In less than five years the same capacity will probably be
available for < $4-500K. Right now, at $2 mill - undergrads have
access to a supercomputer. In a few years, so will probably high
school students.

Look at the list. Non-technical government computers are not the
power-boxes one imagines. Compared to JPL and the national labs,
NSA has never been known for it's ability to design high-tech
software.

Relatively speaking, computing capacity to enhance images is dirt
cheap. Granted, the images on the TV program are doctored (re:
resolution /speed ) - but it's not that hard. Multiple angles, etc.
would be nice to increase probability - but we're not looking to read
license plates from space here. If 'run-of-the-mill' space images
have resolutions < a few meters, images taken from a few dozen yards
away are not all difficult to artifically enhance with varying degrees
of probability.
Anonymous
March 28, 2005 5:33:43 AM

Archived from groups: rec.photo.digital (More info?)

cbrs@silicondairy.net wrote:
> . For now little can be done short
>
>>of governmental budget figures.
>
>
> Can't agree with that....
>
> According to the latest (Nov 2004) rating of the world's 500 fastest
> supercomputers - (see "Top 500 Supercomputer Sites") - #7 is the
> 'home built' machine at Virginia Tech.
> 1100 Apple G5 machines as a MPC. An absolute bargain at
> $2,060,000.00! ($2 mil for the machines / 60K to put it together).
> Bigger bang for the buck than dozens of lesser rated machines at
> government labs, nasa, etc. - and at a school that isn't in the same
> league as MIT, CalTech, etc. with all their government programs
> /support. In less than five years the same capacity will probably be
> available for < $4-500K. Right now, at $2 mill - undergrads have
> access to a supercomputer. In a few years, so will probably high
> school students.
>
> Look at the list. Non-technical government computers are not the
> power-boxes one imagines. Compared to JPL and the national labs,
> NSA has never been known for it's ability to design high-tech
> software.
>
> Relatively speaking, computing capacity to enhance images is dirt
> cheap. Granted, the images on the TV program are doctored (re:
> resolution /speed ) - but it's not that hard. Multiple angles, etc.
> would be nice to increase probability - but we're not looking to read
> license plates from space here. If 'run-of-the-mill' space images
> have resolutions < a few meters, images taken from a few dozen yards
> away are not all difficult to artifically enhance with varying degrees
> of probability.
>
>

If you want to believe you can get blood from a turnip, go ahead and
believe that. But enhancing a lousy surveillance camera image to show a
license plate at 100 yards is NOT going to happen. You MIGHT be able to
guess the make and model of the car, however.



--
Ron Hunter rphunter@charter.net
March 28, 2005 7:03:56 AM

Archived from groups: rec.photo.digital (More info?)

"Mtco" <news@mtco.com> wrote in news:42475929$1_2@127.0.0.1:

> Watching TV shows such as CSI they will Zoom in and make picture
> clear enough to read licensev plate numbers how do they do that?

On TV they can take a digital image where the license plate is 1 pixel high
and 4 pixels wide, they can zoom in and enhance and it looks like a clear
sharp high res image of the plate.

In real life you cannot read a plate when you only have 4 pixels total.

Some of us understand that TV is not 100% realistic.


--
Mark Heyes (New Zealand)
See my pics at www.gigatech.co.nz (last updated 20-Jan-05)
"There are 10 types of people, those that
understand binary and those that don't"
Anonymous
March 28, 2005 7:30:13 AM

Archived from groups: rec.photo.digital (More info?)

"MarkH" <markat@atdot.dot.dot> wrote in message
news:wuK1e.2633884$Zm5.432828@news.easynews.com...
> "Mtco" <news@mtco.com> wrote in news:42475929$1_2@127.0.0.1:
>
>> Watching TV shows such as CSI they will Zoom in and make picture
>> clear enough to read licensev plate numbers how do they do that?
>
> On TV they can take a digital image where the license plate is 1 pixel
> high
> and 4 pixels wide, they can zoom in and enhance and it looks like a clear
> sharp high res image of the plate.
>
> In real life you cannot read a plate when you only have 4 pixels total.
>
> Some of us understand that TV is not 100% realistic.
>
>
> --
> Mark Heyes (New Zealand)
> See my pics at www.gigatech.co.nz (last updated 20-Jan-05)
> "There are 10 types of people, those that
> understand binary and those that don't"

Exactly right Mark, but tell that to Americans and they wont believe you,
LOL. Bottom live is you can't make up what is not there.
Anonymous
March 28, 2005 7:31:25 AM

Archived from groups: rec.photo.digital (More info?)

Well yes you can but you need something like the Gigapixel camera.

http://www.tpd.tno.nl/smartsite966.html

"Mtco" <news@mtco.com> wrote in message news:42475929$1_2@127.0.0.1...
> Watching TV shows such as CSI they will Zoom in and make picture clear
> enough to read licensev plate numbers how do they do that?
>
> Thanks
> Mark
>
>
> ----== Posted via Newsfeeds.Com - Unlimited-Uncensored-Secure Usenet
> News==----
> http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+
> Newsgroups
> ----= East and West-Coast Server Farms - Total Privacy via Encryption
> =----
Anonymous
March 28, 2005 8:16:04 AM

Archived from groups: rec.photo.digital (More info?)

"Mtco" <news@mtco.com> wrote in message news:42475929$1_2@127.0.0.1...
> Watching TV shows such as CSI they will Zoom in and make picture clear
> enough to read licensev plate numbers how do they do that?

It's television.
Anonymous
March 28, 2005 8:43:49 AM

Archived from groups: rec.photo.digital (More info?)

Mtco <news@mtco.com> wrote:
: Are you saying it can't be done?

Not in the way that the TV shows portray it. In many interviews it has
been revealed for this bit of stage trickery they simply start with a
clear photo and apply a mosaic filter to it. The second image is then
portrayed as the original. A simple fade to the original (clear) image
makes the scene work.

There are ways to bring out more detail from a photo to a certain extent.
But the dramatic results as shown on the TV shows is WAY beyond the
current capabilities of desk top computers. But even the most advanced
system can not bring a readable license number out of a single pixel on a
digital photo.

Now the ability of certain individuals to fly without a plane as shown on
TV has to be real. Right? :) 

Randy

==========
Randy Berbaum
Champaign, IL
March 28, 2005 10:08:18 AM

Archived from groups: rec.photo.digital (More info?)

"Mtco" <news@mtco.com> wrote in news:42475929$1_2@127.0.0.1:

> Watching TV shows such as CSI they will Zoom in and make picture
> clear enough to read licensev plate numbers how do they do that?

It would be good if someone could explain how all cases are solved in one
work shift (8-12 hours). In real life it is normal for a case to take
weeks to solve and some take much longer.

To do the digital image manipulation you need to first of all have a PC
running Hollywood OS, then you need a magical image manipulation program
that can create data that was never captured, then . . .


--
Mark Heyes (New Zealand)
See my pics at www.gigatech.co.nz (last updated 20-Jan-05)
"There are 10 types of people, those that
understand binary and those that don't"
Anonymous
March 28, 2005 5:34:19 PM

Archived from groups: rec.photo.digital (More info?)

On Mon, 28 Mar 2005 01:33:40 GMT, in rec.photo.digital ,
"Toomanyputters" <rainydays@theswamp.com> in
<U9J1e.70661$_i3.3942533@twister.southeast.rr.com> wrote:

>
>"Frank ess" <frank@fshe2fs.com> wrote in message
>news:ypednU8aA7pSwNrfRVn-ig@giganews.com...
>> Mtco wrote:
>>> Watching TV shows such as CSI they will Zoom in and make picture
>>> clear enough to read licensev plate numbers how do they do that?
>>>
>>
>> Short answer: They don't. It's Science Fiction. Beam me outta there.
>>
>//snip//
>
>And you know this because?
>
Which is the response I get from my spouse when I snort during
CSI/etc. I love those kinds of shows, but every time they cover some
sci/tech that I know about they get it at least a bit, and often a
very large bit, wrong.

As for my answer to your question, see Ron's answer.


--
Matt Silberstein

All in all, if I could be any animal, I would want to be
a duck or a goose. They can fly, walk, and swim. Plus,
there there is a certain satisfaction knowing that at the
end of your life you will taste good with an orange sauce
or, in the case of a goose, a chestnut stuffing.
Anonymous
March 28, 2005 6:38:04 PM

Archived from groups: rec.photo.digital (More info?)

On Mon, 28 Mar 2005 01:33:40 GMT, "Toomanyputters"
<rainydays@theswamp.com> wrote:

>
>"Frank ess" <frank@fshe2fs.com> wrote in message
>news:ypednU8aA7pSwNrfRVn-ig@giganews.com...
>> Mtco wrote:
>>> Watching TV shows such as CSI they will Zoom in and make picture
>>> clear enough to read licensev plate numbers how do they do that?
>>>
>>
>> Short answer: They don't. It's Science Fiction. Beam me outta there.
>>
>//snip//
>
>And you know this because?
>
Probably becasue he's closely related to reality! :-)
Seriously, it's because we just can't get data where there isn't any.
Trying to read a license plate when the characters can't be made out
is impossible. Even the best of approximating algorythms are only
guessing. For foliage, that may be acceptable, but for alphanumeric
characters, it's not.
For now (and the forseeable future), such capabilities are only within
the realm of imagination.

--
Bill Funk
Change "g" to "a"
Anonymous
March 28, 2005 6:41:03 PM

Archived from groups: rec.photo.digital (More info?)

On Sun, 27 Mar 2005 22:09:25 -0500, cbrs@silicondairy.net wrote:

>Relatively speaking, computing capacity to enhance images is dirt
>cheap. Granted, the images on the TV program are doctored (re:
>resolution /speed ) - but it's not that hard. Multiple angles, etc.
>would be nice to increase probability - but we're not looking to read
>license plates from space here. If 'run-of-the-mill' space images
>have resolutions < a few meters, images taken from a few dozen yards
>away are not all difficult to artifically enhance with varying degrees
>of probability.

Do you have anything other than your sayso to back this up? No one
else does.
If the data isn't in a photo file, it can't be added.
You can use some techniques to make it *look* like detail was added,
but that *added* detail is fake, and can't be used to say, "Look! This
is a licence number that couldn't be seen before!"

--
Bill Funk
Change "g" to "a"
Anonymous
March 28, 2005 11:54:00 PM

Archived from groups: rec.photo.digital (More info?)

Big Bill wrote:
> On Sun, 27 Mar 2005 22:09:25 -0500, cbrs@silicondairy.net wrote:
>
>
>>Relatively speaking, computing capacity to enhance images is dirt
>>cheap. Granted, the images on the TV program are doctored (re:
>>resolution /speed ) - but it's not that hard. Multiple angles, etc.
>>would be nice to increase probability - but we're not looking to read
>>license plates from space here. If 'run-of-the-mill' space images
>>have resolutions < a few meters, images taken from a few dozen yards
>>away are not all difficult to artifically enhance with varying degrees
>>of probability.
>
>
> Do you have anything other than your sayso to back this up? No one
> else does.
> If the data isn't in a photo file, it can't be added.
> You can use some techniques to make it *look* like detail was added,
> but that *added* detail is fake, and can't be used to say, "Look! This
> is a licence number that couldn't be seen before!"
>
While there is often much detail in a digital photo that can be brought
out, and enhanced with editing techniques available to everyone, getting
a clear shot of a car plate, or a face, from a fuzzy VHS videotape from
a surveillance camera is just plain NOT POSSIBLE.


--
Ron Hunter rphunter@charter.net
Anonymous
March 29, 2005 12:16:44 AM

Archived from groups: rec.photo.digital (More info?)

>
>Do you have anything other than your sayso to back this up? No one
>else does.


I'd try the Journal of Electronic Imaging / The Society for Imaging
Science and Technology for starts, then go onto some of the academic
/R&D sites.

http://www.focusmagic.com/ is an interesting site re: capabilities of
even consumer-level / cheap software. Vivid before /after shots
of license plates - and for only $45.00. Gee, Imagine what
'professional /forensic' software might do - the people at the
national labs, and NIST are more into this than one might think.
not to mention corporate R&D labs who find synergy in the technology
across many scientific /technical fields. One company has actually
marketed software to search across security videos for specific people
and groups. No doubt real time evaluation of actual video
surveillance feeds is also possible.

http://csdl.computer.org/comp/proceedings/cvpr/1997/782...
is a summary of an interesting paper by some people at Siemens Corp.
entitled "Character Extraction of License Plates from Video."
Dated, it still demonstrates that professional algorithm software can
do some pretty amazing things. Siemens is pretty good at it, they
have some truly amazing algorithms /software for enhancing medical
images.
An on-line topic seach similar papers reveals dozens of similar
papers.

Current, consumer-available space imagery is at the level of 0.5
meters (i.e. an object +/- 20" can be resolved from space). Congress
has approved its access by the general public, which probably means
government systems can do better. (see mapmart.com - which offers high
resolution - 24" - of the entire world to the public).

Another point I question is "producing images from information that is
not there..." Since the majority of in-place sources for such raw
information are still analog - then the full spectrum of information
is indeed available... the question is how to resolve it accurately
through the use of properly designed sampling and algorithms.
Anonymous
March 29, 2005 1:19:50 AM

Archived from groups: rec.photo.digital (More info?)

On Mon, 28 Mar 2005 19:54:00 -0600, Ron Hunter <rphunter@charter.net>
wrote:

>Big Bill wrote:
>> On Sun, 27 Mar 2005 22:09:25 -0500, cbrs@silicondairy.net wrote:
>>
>>
>>>Relatively speaking, computing capacity to enhance images is dirt
>>>cheap. Granted, the images on the TV program are doctored (re:
>>>resolution /speed ) - but it's not that hard. Multiple angles, etc.
>>>would be nice to increase probability - but we're not looking to read
>>>license plates from space here. If 'run-of-the-mill' space images
>>>have resolutions < a few meters, images taken from a few dozen yards
>>>away are not all difficult to artifically enhance with varying degrees
>>>of probability.
>>
>>
>> Do you have anything other than your sayso to back this up? No one
>> else does.
>> If the data isn't in a photo file, it can't be added.
>> You can use some techniques to make it *look* like detail was added,
>> but that *added* detail is fake, and can't be used to say, "Look! This
>> is a licence number that couldn't be seen before!"
>>
>While there is often much detail in a digital photo that can be brought
>out, and enhanced with editing techniques available to everyone, getting
>a clear shot of a car plate, or a face, from a fuzzy VHS videotape from
>a surveillance camera is just plain NOT POSSIBLE.

You know it, I know it, but cbrs seems to think data that isn't there
can somehow be added accurately.

--
Bill Funk
Change "g" to "a"
Anonymous
March 29, 2005 4:09:57 AM

Archived from groups: rec.photo.digital (More info?)

Big Bill wrote:
> On Mon, 28 Mar 2005 19:54:00 -0600, Ron Hunter <rphunter@charter.net>
> wrote:
>
>
>>Big Bill wrote:
>>
>>>On Sun, 27 Mar 2005 22:09:25 -0500, cbrs@silicondairy.net wrote:
>>>
>>>
>>>
>>>>Relatively speaking, computing capacity to enhance images is dirt
>>>>cheap. Granted, the images on the TV program are doctored (re:
>>>>resolution /speed ) - but it's not that hard. Multiple angles, etc.
>>>>would be nice to increase probability - but we're not looking to read
>>>>license plates from space here. If 'run-of-the-mill' space images
>>>>have resolutions < a few meters, images taken from a few dozen yards
>>>>away are not all difficult to artifically enhance with varying degrees
>>>>of probability.
>>>
>>>
>>>Do you have anything other than your sayso to back this up? No one
>>>else does.
>>>If the data isn't in a photo file, it can't be added.
>>>You can use some techniques to make it *look* like detail was added,
>>>but that *added* detail is fake, and can't be used to say, "Look! This
>>>is a licence number that couldn't be seen before!"
>>>
>>
>>While there is often much detail in a digital photo that can be brought
>>out, and enhanced with editing techniques available to everyone, getting
>>a clear shot of a car plate, or a face, from a fuzzy VHS videotape from
>>a surveillance camera is just plain NOT POSSIBLE.
>
>
> You know it, I know it, but cbrs seems to think data that isn't there
> can somehow be added accurately.
>

Can't you hear the court discussion:

Witness: I used this amazing fractal program for clarifying the license
number.

Attorney: And how does this program work?

Witness: Well, if looks at the available data and then invents new
pixels consistent with the available ones to restore the image.

Attorney: So it makes up new pixels?

Witness: Well, yes.

Judge: Case dismissed.


--
Ron Hunter rphunter@charter.net
March 29, 2005 6:51:20 AM

Archived from groups: rec.photo.digital (More info?)

On Sun, 27 Mar 2005 22:09:25 -0500
In message <o4se41tn2e95cob3torunn61ev2j022r19@4ax.com>
Posted from Posted via Supernews, http://www.supernews.com
cbrs@silicondairy.net wrote:

> . For now little can be done short
> > of governmental budget figures.
>
> Can't agree with that....
>
> According to the latest (Nov 2004) rating of the world's 500 fastest
> supercomputers - (see "Top 500 Supercomputer Sites") - #7 is the
> 'home built' machine at Virginia Tech.

(side note: A tech university is not a home. Nevertheless...)

> 1100 Apple G5 machines as a MPC. An absolute bargain at
> $2,060,000.00!
> <SNIP>

Heck, since they are *that* cheap, order me TWO!

Then I'll hire Adobe for a few years to program the thing...

Jeff
March 29, 2005 7:12:00 AM

Archived from groups: rec.photo.digital (More info?)

On Sun, 27 Mar 2005 19:41:25 -0600
In message <42476177$1_1@127.0.0.1>
Posted from Newsfeeds.com
"Mtco" <news@mtco.com> wrote:

> Are you saying it can't be done?

Heck, anyone can watch TV, sheesh. No, wait!

You're saying a program exists that transforms a 640x480 video tape
frame into a 100% accurate 640000 x 480000 tack sharp image.

Put your money where your mouth is...show us proof. The computing
world would salivate at a non-key decompression method that can turn a
5 bit image into a readable license plate. A fixed 50:1 lossless
compression method would be worth a fortune.

.....OR....

DON'T FEED THE TROLL ;^)

Jeff
March 29, 2005 7:24:01 AM

Archived from groups: rec.photo.digital (More info?)

On Mon, 28 Mar 2005 03:30:13 GMT
In message <9TK1e.15009$C7.2216@news-server.bigpond.net.au>
Posted from BigPond Internet Services
"Pete D" <no@email.com> wrote:

> Exactly right Mark, but tell that to Americans and
> they wont believe you, LOL. Bottom live is you can't
> make up what is not there.

Whoosh! Ya'ponders missed your mark by a Hollywood Mile.

The film/video trick is done by starting with a high resolution image
and then shrinking down to the target starting size and noise level.
The results are then presented backwards for production.

Jeff
Anonymous
March 29, 2005 11:25:27 AM

Archived from groups: rec.photo.digital (More info?)

[A complimentary Cc of this posting was sent to
Big Bill
<bill@pipping.com>], who wrote in article <nolh41568e79jb17tull5om61fnaqi6ub3@4ax.com>:
> >While there is often much detail in a digital photo that can be brought
> >out, and enhanced with editing techniques available to everyone, getting
> >a clear shot of a car plate, or a face, from a fuzzy VHS videotape from
> >a surveillance camera is just plain NOT POSSIBLE.
>
> You know it, I know it, but cbrs seems to think data that isn't there
> can somehow be added accurately.

Assuming the car stays there for several minutes, you can almost
completely eliminate the noise by averaging. So you need at least
about 2 optically distinguished pixels per digit (more per letter...);
or about 100 pixels per 10m. Again, due to virtually no noise in the
image, you can go practically to extinction resolution of the lense,
or of the sensor (whichever is less).

I do not know the relation of extinction resolution to the
"practically visible" one for surveilance camera images written on
analogue media; probably no more than 1.5 times difference. So with
practically visible resolution of about 350 pixels (is it?) you may
have a chance to distinguish the number if the field of view is up to
50m. [Correct these numbers accordingly for different types of
surveillance cameras...]

You will not "see" the digits at such a low resolution, but still you
may be able to say "these 6 pixels fit 8 much better than 1".

Hope this helps,
Ilya
Anonymous
March 29, 2005 12:34:12 PM

Archived from groups: rec.photo.digital (More info?)

Ilya Zakharevich wrote:
> [A complimentary Cc of this posting was sent to
> Big Bill
> <bill@pipping.com>], who wrote in article <nolh41568e79jb17tull5om61fnaqi6ub3@4ax.com>:
>
>>>While there is often much detail in a digital photo that can be brought
>>>out, and enhanced with editing techniques available to everyone, getting
>>>a clear shot of a car plate, or a face, from a fuzzy VHS videotape from
>>>a surveillance camera is just plain NOT POSSIBLE.
>>
>>You know it, I know it, but cbrs seems to think data that isn't there
>>can somehow be added accurately.
>
>
> Assuming the car stays there for several minutes, you can almost
> completely eliminate the noise by averaging. So you need at least
> about 2 optically distinguished pixels per digit (more per letter...);
> or about 100 pixels per 10m. Again, due to virtually no noise in the
> image, you can go practically to extinction resolution of the lense,
> or of the sensor (whichever is less).
>
> I do not know the relation of extinction resolution to the
> "practically visible" one for surveilance camera images written on
> analogue media; probably no more than 1.5 times difference. So with
> practically visible resolution of about 350 pixels (is it?) you may
> have a chance to distinguish the number if the field of view is up to
> 50m. [Correct these numbers accordingly for different types of
> surveillance cameras...]
>
> You will not "see" the digits at such a low resolution, but still you
> may be able to say "these 6 pixels fit 8 much better than 1".
>
> Hope this helps,
> Ilya
Yes, but then try to convince me you can distinguish an 8 from a 0 or an
'O' from a 'Q'. Not feasible.


--
Ron Hunter rphunter@charter.net
Anonymous
March 29, 2005 2:34:25 PM

Archived from groups: rec.photo.digital (More info?)

On Tue, 29 Mar 2005 07:25:27 +0000 (UTC), Ilya Zakharevich
<nospam-abuse@ilyaz.org> wrote:

>[A complimentary Cc of this posting was sent to
>Big Bill
><bill@pipping.com>], who wrote in article <nolh41568e79jb17tull5om61fnaqi6ub3@4ax.com>:
>> >While there is often much detail in a digital photo that can be brought
>> >out, and enhanced with editing techniques available to everyone, getting
>> >a clear shot of a car plate, or a face, from a fuzzy VHS videotape from
>> >a surveillance camera is just plain NOT POSSIBLE.
>>
>> You know it, I know it, but cbrs seems to think data that isn't there
>> can somehow be added accurately.
>
>Assuming the car stays there for several minutes, you can almost
>completely eliminate the noise by averaging. So you need at least
>about 2 optically distinguished pixels per digit (more per letter...);
>or about 100 pixels per 10m. Again, due to virtually no noise in the
>image, you can go practically to extinction resolution of the lense,
>or of the sensor (whichever is less).

If the pixels are of a size that a character on the plate is only
coverd by one or two pixels, therer is no way the character can be
discerend.
Especially the way the OP is talkig about - from a single image.
Yes, I'm aware that, iusing a lot of images from different angles, a
close approximation can be made, but that's all it is: a close
approximation.
Would you convict on such an approximation? Would you like to be
convicted on such an approximation?
>
>I do not know the relation of extinction resolution to the
>"practically visible" one for surveilance camera images written on
>analogue media; probably no more than 1.5 times difference. So with
>practically visible resolution of about 350 pixels (is it?) you may
>have a chance to distinguish the number if the field of view is up to
>50m. [Correct these numbers accordingly for different types of
>surveillance cameras...]
>
>You will not "see" the digits at such a low resolution, but still you
>may be able to say "these 6 pixels fit 8 much better than 1".

But do they fit "B" better than "8"?
I've seen the scenarios the OP is asking about (Las Vegas, anyone?).
We aren't talking about 6 pixeld per character, we're talking about a
face over a half mile away using what appears to be a standard
surveilance camera, outdoors (that means the lens/covering is subject
to whatever environmental smutz that gets on it), zoomed to the max,
with the face barely recognizable (and wouldn't be, except for the
fact that it sits on top of what looks like a body), that is enhanced
to bring the face into sharp focus.
Impossible with today's technology.

>
>Hope this helps,
>Ilya

--
Bill Funk
Change "g" to "a"
Anonymous
March 29, 2005 5:28:25 PM

Archived from groups: rec.photo.digital (More info?)

In article <42475929$1_2@127.0.0.1>, Mtco <news@mtco.com> wrote:
>Watching TV shows such as CSI they will Zoom in and make picture clear
>enough to read licensev plate numbers how do they do that?

You use a magic computer running Hollywood OS(tm). It has a feature whereby
it can resolve unreadable details in images simply by the operator issuing
the (verbal) command, "enhance!", and can also crack passwords one character
at a time, but has the critical flaw that it only prints text at the speed
of a daisy wheel printer and must beep every time a character appears. It
also only allows you to save one copy of any file, ever.
Anonymous
March 29, 2005 7:07:07 PM

Archived from groups: rec.photo.digital (More info?)

On Tue, 29 Mar 2005 08:34:12 -0600, Ron Hunter <rphunter@charter.net>
wrote:

>Ilya Zakharevich wrote:
>
>> Assuming the car stays there for several minutes, you can almost
>> completely eliminate the noise by averaging. So you need at least
>> about 2 optically distinguished pixels per digit (more per letter...);
>> or about 100 pixels per 10m. Again, due to virtually no noise in the
>> image, you can go practically to extinction resolution of the lense,
>> or of the sensor (whichever is less).
>>
>> I do not know the relation of extinction resolution to the
>> "practically visible" one for surveilance camera images written on
>> analogue media; probably no more than 1.5 times difference. So with
>> practically visible resolution of about 350 pixels (is it?) you may
>> have a chance to distinguish the number if the field of view is up to
>> 50m. [Correct these numbers accordingly for different types of
>> surveillance cameras...]
>>
>> You will not "see" the digits at such a low resolution, but still you
>> may be able to say "these 6 pixels fit 8 much better than 1".
>>
>> Hope this helps,
>> Ilya
>Yes, but then try to convince me you can distinguish an 8 from a 0 or an
>'O' from a 'Q'. Not feasible.

Run the plate through the licensing authorities.

ABC Q8 = No hits.
ABC O8 = One hit. Blue Chevvy, Nebraska.
ABC Q0 = Two hits. Silver Honda, Florida. Green Toyota, Alaska.
ABC O0 = No hits.

Now tell me you can't read the plate.

It all comes down to probabilities.

--
Owamanga!
http://www.pbase.com/owamanga
Anonymous
March 29, 2005 7:07:08 PM

Archived from groups: rec.photo.digital (More info?)

Owamanga wrote:
> On Tue, 29 Mar 2005 08:34:12 -0600, Ron Hunter <rphunter@charter.net>
> wrote:
>
>
>>Ilya Zakharevich wrote:
>>
>>
>>>Assuming the car stays there for several minutes, you can almost
>>>completely eliminate the noise by averaging. So you need at least
>>>about 2 optically distinguished pixels per digit (more per letter...);
>>>or about 100 pixels per 10m. Again, due to virtually no noise in the
>>>image, you can go practically to extinction resolution of the lense,
>>>or of the sensor (whichever is less).
>>>
>>>I do not know the relation of extinction resolution to the
>>>"practically visible" one for surveilance camera images written on
>>>analogue media; probably no more than 1.5 times difference. So with
>>>practically visible resolution of about 350 pixels (is it?) you may
>>>have a chance to distinguish the number if the field of view is up to
>>>50m. [Correct these numbers accordingly for different types of
>>>surveillance cameras...]
>>>
>>>You will not "see" the digits at such a low resolution, but still you
>>>may be able to say "these 6 pixels fit 8 much better than 1".
>>>
>>>Hope this helps,
>>>Ilya
>>
>>Yes, but then try to convince me you can distinguish an 8 from a 0 or an
>>'O' from a 'Q'. Not feasible.
>
>
> Run the plate through the licensing authorities.
>
> ABC Q8 = No hits.
> ABC O8 = One hit. Blue Chevvy, Nebraska.
> ABC Q0 = Two hits. Silver Honda, Florida. Green Toyota, Alaska.
> ABC O0 = No hits.
>
> Now tell me you can't read the plate.
>
> It all comes down to probabilities.
>
> --
> Owamanga!
> http://www.pbase.com/owamanga

And if the plate seems to contain several 'O', '0', and '8' characters?
Now make sure you have the right state...


--
Ron Hunter rphunter@charter.net
Anonymous
March 29, 2005 7:07:08 PM

Archived from groups: rec.photo.digital (More info?)

On Tue, 29 Mar 2005 15:07:07 GMT, Owamanga
<owamanga(not-this-bit)@hotmail.com> wrote:

>>Yes, but then try to convince me you can distinguish an 8 from a 0 or an
>>'O' from a 'Q'. Not feasible.
>
>Run the plate through the licensing authorities.
>
>ABC Q8 = No hits.
>ABC O8 = One hit. Blue Chevvy, Nebraska.
>ABC Q0 = Two hits. Silver Honda, Florida. Green Toyota, Alaska.
>ABC O0 = No hits.
>
>Now tell me you can't read the plate.
>
>It all comes down to probabilities.

Ah yes, another hypothetical situation, where you get to set the
parameters.
Where did the "ABC" come from? What sets "ABC" apart from "A8O"? Or
"NOO"? Or any of hundreds of other possibilities that are every bit as
ambiguous as Q8/Q0/O8/O0?
The possibilities quickly expand to a very large number, with
consequently more hits. And a consequently lower probability of
success.
Which means a much lower probability of an accurate description of
anything.
--
Bill Funk
Change "g" to "a"
March 29, 2005 7:07:09 PM

Archived from groups: rec.photo.digital (More info?)

In article <od4j41trt6ckejrjq9tl7o6n0rk7m8p18f@4ax.com>, bill@pipping.com
says...
> Ah yes, another hypothetical situation, where you get to set the
> parameters.
> Where did the "ABC" come from? What sets "ABC" apart from "A8O"? Or
> "NOO"? Or any of hundreds of other possibilities that are every bit as
> ambiguous as Q8/Q0/O8/O0?
> The possibilities quickly expand to a very large number, with
> consequently more hits. And a consequently lower probability of
> success.
> Which means a much lower probability of an accurate description of
> anything.
> --
> Bill Funk
> Change "g" to "a"
>
>


If you know what state or group of state are likely to be involved you can
eliminat a lot of possibilities.

Most (maybe all) states have combinations they DO NOT USE on certain
vehicles, and other combinations that are only used on certain vehicles,
and combinations that are only used for the first (or last) three digits.


--
Larry Lynch
Mystic, Ct.
Anonymous
March 29, 2005 8:15:27 PM

Archived from groups: rec.photo.digital (More info?)

On Tue, 29 Mar 2005 15:07:07 GMT, in rec.photo.digital , Owamanga
<owamanga(not-this-bit)@hotmail.com> in
<1tri41d88got4v8jk964paod49m3n28dpg@4ax.com> wrote:

>On Tue, 29 Mar 2005 08:34:12 -0600, Ron Hunter <rphunter@charter.net>
>wrote:
>
>>Ilya Zakharevich wrote:
>>
>>> Assuming the car stays there for several minutes, you can almost
>>> completely eliminate the noise by averaging. So you need at least
>>> about 2 optically distinguished pixels per digit (more per letter...);
>>> or about 100 pixels per 10m. Again, due to virtually no noise in the
>>> image, you can go practically to extinction resolution of the lense,
>>> or of the sensor (whichever is less).
>>>
>>> I do not know the relation of extinction resolution to the
>>> "practically visible" one for surveilance camera images written on
>>> analogue media; probably no more than 1.5 times difference. So with
>>> practically visible resolution of about 350 pixels (is it?) you may
>>> have a chance to distinguish the number if the field of view is up to
>>> 50m. [Correct these numbers accordingly for different types of
>>> surveillance cameras...]
>>>
>>> You will not "see" the digits at such a low resolution, but still you
>>> may be able to say "these 6 pixels fit 8 much better than 1".
>>>
>>> Hope this helps,
>>> Ilya
>>Yes, but then try to convince me you can distinguish an 8 from a 0 or an
>>'O' from a 'Q'. Not feasible.
>
>Run the plate through the licensing authorities.
>
>ABC Q8 = No hits.
>ABC O8 = One hit. Blue Chevvy, Nebraska.
>ABC Q0 = Two hits. Silver Honda, Florida. Green Toyota, Alaska.
>ABC O0 = No hits.
>
>Now tell me you can't read the plate.
>
>It all comes down to probabilities.

Color will tell you state. Try again.



--
Matt Silberstein

All in all, if I could be any animal, I would want to be
a duck or a goose. They can fly, walk, and swim. Plus,
there there is a certain satisfaction knowing that at the
end of your life you will taste good with an orange sauce
or, in the case of a goose, a chestnut stuffing.
Anonymous
March 29, 2005 8:15:28 PM

Archived from groups: rec.photo.digital (More info?)

Matt Silberstein wrote:
> On Tue, 29 Mar 2005 15:07:07 GMT, in rec.photo.digital , Owamanga
> <owamanga(not-this-bit)@hotmail.com> in
> <1tri41d88got4v8jk964paod49m3n28dpg@4ax.com> wrote:
>
>
>>On Tue, 29 Mar 2005 08:34:12 -0600, Ron Hunter <rphunter@charter.net>
>>wrote:
>>
>>
>>>Ilya Zakharevich wrote:
>>>
>>>
>>>>Assuming the car stays there for several minutes, you can almost
>>>>completely eliminate the noise by averaging. So you need at least
>>>>about 2 optically distinguished pixels per digit (more per letter...);
>>>>or about 100 pixels per 10m. Again, due to virtually no noise in the
>>>>image, you can go practically to extinction resolution of the lense,
>>>>or of the sensor (whichever is less).
>>>>
>>>>I do not know the relation of extinction resolution to the
>>>>"practically visible" one for surveilance camera images written on
>>>>analogue media; probably no more than 1.5 times difference. So with
>>>>practically visible resolution of about 350 pixels (is it?) you may
>>>>have a chance to distinguish the number if the field of view is up to
>>>>50m. [Correct these numbers accordingly for different types of
>>>>surveillance cameras...]
>>>>
>>>>You will not "see" the digits at such a low resolution, but still you
>>>>may be able to say "these 6 pixels fit 8 much better than 1".
>>>>
>>>>Hope this helps,
>>>>Ilya
>>>
>>>Yes, but then try to convince me you can distinguish an 8 from a 0 or an
>>>'O' from a 'Q'. Not feasible.
>>
>>Run the plate through the licensing authorities.
>>
>>ABC Q8 = No hits.
>>ABC O8 = One hit. Blue Chevvy, Nebraska.
>>ABC Q0 = Two hits. Silver Honda, Florida. Green Toyota, Alaska.
>>ABC O0 = No hits.
>>
>>Now tell me you can't read the plate.
>>
>>It all comes down to probabilities.
>
>
> Color will tell you state. Try again.
>
>
>
Color is VERY hard to distinguish from a B&W surveillance video!


--
Ron Hunter rphunter@charter.net
Anonymous
March 29, 2005 8:44:30 PM

Archived from groups: rec.photo.digital (More info?)

On Tue, 29 Mar 2005 16:15:27 GMT, Matt Silberstein
<RemoveThisPrefixmatts2nospam@ix.netcom.com> wrote:

>On Tue, 29 Mar 2005 15:07:07 GMT, in rec.photo.digital , Owamanga
><owamanga(not-this-bit)@hotmail.com> in
><1tri41d88got4v8jk964paod49m3n28dpg@4ax.com> wrote:
>
>>On Tue, 29 Mar 2005 08:34:12 -0600, Ron Hunter <rphunter@charter.net>
>>wrote:
>>
>>>Ilya Zakharevich wrote:
>>>
>>>> Assuming the car stays there for several minutes, you can almost
>>>> completely eliminate the noise by averaging. So you need at least
>>>> about 2 optically distinguished pixels per digit (more per letter...);
>>>> or about 100 pixels per 10m. Again, due to virtually no noise in the
>>>> image, you can go practically to extinction resolution of the lense,
>>>> or of the sensor (whichever is less).
>>>>
>>>> I do not know the relation of extinction resolution to the
>>>> "practically visible" one for surveilance camera images written on
>>>> analogue media; probably no more than 1.5 times difference. So with
>>>> practically visible resolution of about 350 pixels (is it?) you may
>>>> have a chance to distinguish the number if the field of view is up to
>>>> 50m. [Correct these numbers accordingly for different types of
>>>> surveillance cameras...]
>>>>
>>>> You will not "see" the digits at such a low resolution, but still you
>>>> may be able to say "these 6 pixels fit 8 much better than 1".
>>>>
>>>> Hope this helps,
>>>> Ilya
>>>Yes, but then try to convince me you can distinguish an 8 from a 0 or an
>>>'O' from a 'Q'. Not feasible.
>>
>>Run the plate through the licensing authorities.
>>
>>ABC Q8 = No hits.
>>ABC O8 = One hit. Blue Chevvy, Nebraska.
>>ABC Q0 = Two hits. Silver Honda, Florida. Green Toyota, Alaska.
>>ABC O0 = No hits.
>>
>>Now tell me you can't read the plate.
>>
>>It all comes down to probabilities.
>
>Color will tell you state. Try again.

Not sure what your point is, each state has over 100 different plate
colors.

Anyway, if the picture is of a silver honda with a Florida plate, then
I can read the plate: ABC Q0, even if I can't see it properly.

--
Owamanga!
http://www.pbase.com/owamanga
Anonymous
March 29, 2005 9:07:53 PM

Archived from groups: rec.photo.digital (More info?)

On Tue, 29 Mar 2005 10:37:52 -0600, Ron Hunter <rphunter@charter.net>
wrote:

>Owamanga wrote:
>> On Tue, 29 Mar 2005 08:34:12 -0600, Ron Hunter <rphunter@charter.net>
>> wrote:
>>
>>>Yes, but then try to convince me you can distinguish an 8 from a 0 or an
>>>'O' from a 'Q'. Not feasible.
>>
>>
>> Run the plate through the licensing authorities.
>>
>> ABC Q8 = No hits.
>> ABC O8 = One hit. Blue Chevvy, Nebraska.
>> ABC Q0 = Two hits. Silver Honda, Florida. Green Toyota, Alaska.
>> ABC O0 = No hits.
>>
>> Now tell me you can't read the plate.
>>
>> It all comes down to probabilities.
>>
>
>And if the plate seems to contain several 'O', '0', and '8' characters?
>Now make sure you have the right state...

Probability again. There are thousands of models of cars which help
identify a given car. Each of those come in many different colors
(some of which will be obvious even on a B&W camera) even if you say
light, medium dark. Finally, the geographical location of the camera
vs the states of which you are getting matches in go into the math and
you end up with a list of 1 or more 'suspects' to check out in more
detail.

--
Owamanga!
http://www.pbase.com/owamanga
Anonymous
March 29, 2005 10:38:32 PM

Archived from groups: rec.photo.digital (More info?)

On Tue, 29 Mar 2005 10:39:23 -0700, Big Bill <bill@pipping.com> wrote:

>On Tue, 29 Mar 2005 15:07:07 GMT, Owamanga
><owamanga(not-this-bit)@hotmail.com> wrote:
>
>>It all comes down to probabilities.
>
>Ah yes, another hypothetical situation, where you get to set the
>parameters.

It was an illustration, and of course I get to set the parameters.

>Where did the "ABC" come from?

First three letters of the alphabet. Why?

>What sets "ABC" apart from "A8O"? Or
>"NOO"? Or any of hundreds of other possibilities that are every bit as
>ambiguous as Q8/Q0/O8/O0?

Nothing, I just chose them for illustration.

>The possibilities quickly expand to a very large number, with
>consequently more hits. And a consequently lower probability of
>success.

Get a computer to add 10 numbers or 100,000 numbers it makes no
difference. Large numbers are our friends.

Take a sentence, remove 20% of the characters and most people can
still read it. Because we know what's legal english and what isn't.
This same method can be employed by computers who have access to
vehicle license data to rule out irrelevant hits and give a list based
on probability.

"I'x tixed of axl txis nonsexse abxut beaxty xeing only skxn-deep.
Thxt'x deex enoxgh. Wxat dx yox want, an adoxable pxncreas?"

>Which means a much lower probability of an accurate description of
>anything.

Probability gets worse with lower quality pictures, yes, but you'll
still get the answer somewhere in the list. It's just.. how big is
that list going to be?

--
Owamanga!
http://www.pbase.com/owamanga
Anonymous
March 30, 2005 12:01:34 AM

Archived from groups: rec.photo.digital (More info?)

Owamanga wrote:
> On Tue, 29 Mar 2005 10:37:52 -0600, Ron Hunter <rphunter@charter.net>
> wrote:
>
>
>>Owamanga wrote:
>>
>>>On Tue, 29 Mar 2005 08:34:12 -0600, Ron Hunter <rphunter@charter.net>
>>>wrote:
>>>
>>>
>>>>Yes, but then try to convince me you can distinguish an 8 from a 0 or an
>>>>'O' from a 'Q'. Not feasible.
>>>
>>>
>>>Run the plate through the licensing authorities.
>>>
>>>ABC Q8 = No hits.
>>>ABC O8 = One hit. Blue Chevvy, Nebraska.
>>>ABC Q0 = Two hits. Silver Honda, Florida. Green Toyota, Alaska.
>>>ABC O0 = No hits.
>>>
>>>Now tell me you can't read the plate.
>>>
>>>It all comes down to probabilities.
>>>
>>
>>And if the plate seems to contain several 'O', '0', and '8' characters?
>>Now make sure you have the right state...
>
>
> Probability again. There are thousands of models of cars which help
> identify a given car. Each of those come in many different colors
> (some of which will be obvious even on a B&W camera) even if you say
> light, medium dark. Finally, the geographical location of the camera
> vs the states of which you are getting matches in go into the math and
> you end up with a list of 1 or more 'suspects' to check out in more
> detail.
>
> --
> Owamanga!
> http://www.pbase.com/owamanga
Or 10,000, or more.


--
Ron Hunter rphunter@charter.net
Anonymous
March 30, 2005 8:05:35 AM

Archived from groups: rec.photo.digital (More info?)

[A complimentary Cc of this posting was sent to
Big Bill
<bill@pipping.com>], who wrote in article <tt3j41h92cnjn21nne0e3su1far0cotgmm@4ax.com>:
> >Assuming the car stays there for several minutes, you can almost
> >completely eliminate the noise by averaging. So you need at least
> >about 2 optically distinguished pixels per digit (more per letter...);
> >or about 100 pixels per 10m. Again, due to virtually no noise in the
> >image, you can go practically to extinction resolution of the lense,
> >or of the sensor (whichever is less).

> If the pixels are of a size that a character on the plate is only
> coverd by one or two pixels, therer is no way the character can be
> discerend.

If a character has width of 2 pixels, then its height is about 3-4
pixels. This gives you 7 grayscale pixels; this is quite enough.

> Would you convict on such an approximation? Would you like to be
> convicted on such an approximation?

My understanding is most evidence is used as a tool to collect some
other evidence.

> I've seen the scenarios the OP is asking about (Las Vegas, anyone?).
> We aren't talking about 6 pixeld per character, we're talking about a
> face over a half mile away using what appears to be a standard
> surveilance camera, outdoors (that means the lens/covering is subject
> to whatever environmental smutz that gets on it), zoomed to the max,
> with the face barely recognizable (and wouldn't be, except for the
> fact that it sits on top of what looks like a body), that is enhanced

I was discussing a quite concrete scenario. Now you are saying that
some other scenario is bogus. While I agree with you, it has little
relation to what I said. ;-)

Yours,
Ilya
March 30, 2005 9:36:32 AM

Archived from groups: rec.photo.digital (More info?)

On Wed, 30 Mar 2005 04:05:35 +0000 (UTC)
In message <d2d8if$11me$1@agate.berkeley.edu>
Posted from U.C. Berkeley Math. Department.
Ilya Zakharevich <nospam-abuse@ilyaz.org> wrote:

> <snip>
> I was discussing a quite concrete scenario. <SNIP>

Indeed... we were discussing a different and impossible data creation
scenario. You have 5 pixels, re-create a license plate.

Decades ago we learned how to render 2x5 dot letters that were
readable.

Jeff
Anonymous
March 30, 2005 1:52:11 PM

Archived from groups: rec.photo.digital (More info?)

>>>>> "Mtco" == Mtco <news@mtco.com> writes:

Mtco> Watching TV shows such as CSI they will Zoom in and make
Mtco> picture clear enough to read licensev plate numbers how do
Mtco> they do that?

because the camera was a super-high resolution camera? ;-)

I like the CSI episode where they take the video from a standard
consumer video camera, look at the reflection in the persons eye, and
just about identify the photographer in the process (who was standing
behind the camera).

If you look closely though, you will see evidence that these things
are often faked. For example, on an episode of NCIS, the detail in the
image is too dark. So they change the value of grey from black to
white. Or, another words, the image had been deliberately altered
before hand (without pushing OK) to make greys look black.

Along similar lines, I really like it how they can split audio into
multi sources, e.g. an audio track for car 1, another audio track for
car 2, etc. Only problem is if you look closely you will see (at least
in some episodes) that the tracks were already split up before hand,
and probably recorded that way.

Supposedly CSI is meant to be realistic impression of what CSI is
about, I can't help think that there are probably a lot of
discrepancies... For starters, I suspect there would be a lot more
paper work in order to clearly document evidence that would be
considered too boring for a TV show.

I am waiting for the episode where the case falls through because they
can't remember for certain if the fingerprint came from the lamp on
the left side of the bed or the lamp on the right side of the bed
(viewed from which direction?), or the bathroom light switch. ;-)
--
Brian May <bam@snoopy.apana.org.au>
Anonymous
March 30, 2005 1:52:12 PM

Archived from groups: rec.photo.digital (More info?)

On Wed, 30 Mar 2005 09:52:11 +1000, in rec.photo.digital , Brian May
<bam@snoopy.apana.org.au> in <sa43bue9fis.fsf@snoopy.apana.org.au>
wrote:

>>>>>> "Mtco" == Mtco <news@mtco.com> writes:
>
> Mtco> Watching TV shows such as CSI they will Zoom in and make
> Mtco> picture clear enough to read licensev plate numbers how do
> Mtco> they do that?
>
>because the camera was a super-high resolution camera? ;-)
>
>I like the CSI episode where they take the video from a standard
>consumer video camera, look at the reflection in the persons eye, and
>just about identify the photographer in the process (who was standing
>behind the camera).
>
>If you look closely though, you will see evidence that these things
>are often faked. For example, on an episode of NCIS, the detail in the
>image is too dark. So they change the value of grey from black to
>white. Or, another words, the image had been deliberately altered
>before hand (without pushing OK) to make greys look black.

On IMDB in the Goofs section of _Finding Nemo_ someone pointed out
that Australia does not have Brown Pelicans. I wanted to write in
something mentioning that fish can't talk.

It is fiction folks.

[snip]


--
Matt Silberstein

All in all, if I could be any animal, I would want to be
a duck or a goose. They can fly, walk, and swim. Plus,
there there is a certain satisfaction knowing that at the
end of your life you will taste good with an orange sauce
or, in the case of a goose, a chestnut stuffing.
March 30, 2005 1:52:12 PM

Archived from groups: rec.photo.digital (More info?)

In message <sa43bue9fis.fsf@snoopy.apana.org.au>
Brian May <bam@snoopy.apana.org.au> wrote:

> ...
> Along similar lines, I really like it how they can split audio into
> multi sources, e.g. an audio track for car 1, another audio track for
> car 2, etc. Only problem is if you look closely you will see (at least
> in some episodes) that the tracks were already split up before hand,
> and probably recorded that way.
> ...

Obviously, tv/film recorded public safety radio tracks are
orchestrated. In real life, however, multi-channel monitoring systems
exist. Many radio "systems" have radios that are uniquely identified
by en embedded code in the signal. (When the radio transmits the
radio ID is sent with the signal, there may be a control channel,
there may be several non-coded radios using several frequencies.
Regardless how it's implemented, the coordinator knows who is assigned
to each radio and vehicle, conversations can be public or private as
required within a radio system.)

One radio can monitor a fleet of radios, or several radios can
selectively monitor one or more radios. Place them in an arc in front
of a map area on a table (or floor) and you have a very versatile
mobile command center. The same thing can be done with a rack mount
system and a mixing board. There are trade-offs between the two
methods.

CSI depicts a rudimentary radio system; simple to implement with even
a one or two frequency trunking system for a small group like a CSI
team. All you need is government clearance and there ya go!

Jeff
!