Thursday, December 12, 2002

Time for a good rant...

California State Senator Jack Scott, ever impervious to little things
like facts, has proposed href=""
. SB35 is Jack Scott's latest pipe dream:
Ballistic "Fingerprinting". (Or as Scott fantasizes, "DNA for guns".)

Back in October, the politicos in DC were proposing this at the
National level. The NRA, of course, opposed this for a variety of
reasons. Now there are plenty of people who refuse to listen to
anything the NRA says. No matter how accurate they might be, there are
people who automatically ignore whatever they say on any subject.

Will you listen to a shooter and
optical engineer who knows a fair amount about computer science?

Why am I qualified to voice an opinion? Because I have a certain level
expertise in the sciences that figure most prominantly in this issue. I
know guns and I know the optical systems that a ballistic
fingerprinting system would use. I also have a fair amount of knowledge
on computer science. First, let's start with some technical background

A ballistic "fingerprint" is a photographic record of the markings a
firearm leaves on an expended bullet and its discarded shell casing.
Whenever a round is fired, the gun leaves marks on its components. The
barrel's rifling leaves a pattern of grooves on the bullet. The firing
chamber, extractor, and firing pin leave marks on the metallic shell
casing. Thus the components from two rounds fired through two different
guns will have different markings. It is thus style="font-style: italic;">possible to match a particular
bullet to a particular gun.

Once the fired components are recovered, they are photographed to
produce a record of the marks. This isn't easy since the bullets and
casings are cylindrical. The photo must "unwrap" the image by rotating
the component on a special stage while it's being photographed.
Preferably, more than one rotation about the circumfrence is recorded
to allow for offsetting while indexing the photo to other images. The
rear of the casing must also be photographed to record the marks left
by the firing pin.

OK... so what's the problem?

Let's start with the gun. That "fingerprint" isn't a fingerprint at
all. Your fingerprints are a permanent
feature of your body. They'll be a part of your body even after it's
been laid low and you're off to your eternal reward. They do not
change. The marks left by a gun, however, are style="font-weight: bold;">not permanent features of the gun;
they are transient features. Barrels wear as ammunition is fired
through them. Different types of ammo cause different amounts of barrel
wear. Lead bullets (as opposed to copper jacketed bullets) leave
residue in the barrel that fills in smaller defects. Thus the scratches
left by these smaller defects vanish if lead is present. Firing
chambers also wear and brass can fill smaller features thus obscuring
their marks. Firing pins and extractors also wear and, if broken, are
easily replaced. Thus time alone changes the characteristics of the
markings a firearm produces.

It is also very easy to deliberately change the markings. Some shooters
polish the bores of their guns to improve accuracy. Something as simple
as a dab of toothpaste on the tip of a round can alter the barrel's
scratch pattern. Chambers can be similarly polished. (To improve ease
of loading and extraction.) Barrels can also be replaced.

(An added note: In many rifle calibers, it's possible to fire a smaller
bullet with a sabot. The bullet will then be left with only the feeble
markings from the disposable sabot; not the rifle barrel. It is thus
possible to fire a bullet that has style="font-weight: bold;">no fingerprint whatsoever!)

Thus by time or by deliberate effort, a gun's "fingerprint" can be
erased. The photographic record becomes worthless.

But what if the gun is neither worn nor deliberately altered? Could the
record be of use then? Perhaps... If you're lucky.

I mentioned that the images must be recorded on a rotating stage. The
wrinkle there is that there is no correct way to index the bullet or
casing in the stage. There's no up or down to the thing. Neither is
there a correct orientation for the back of the case. Thus the before
and after images are aligned to different rotations. Your image
processing system must first deal with this.

The easiest thing to do would be to take a Fourier transform of the
over-rotated image. This records the spatial distribution of the features
in the image. This distrbution is independent of the indexing. This
provides information that a computer can use more easily to sort out
possible matches to the gun under test. (It will still be necessary to
do a final examination of the candidate matches by hand.) However, not
all of the spatial information is useful as not all of the marks were
left by the gun. New cases jostling together in their box on the way to
the gun store will leave marks on one another. So will the pavement
when the expended case hits the ground. Bone will deform the bullet.
All of these things leave their own spatial signature in the transform.
Buried in this mish-mash of data is the spatial signature of the gun.
An image processing system must be able to discard the false data
without losing the real record. This isn't easy. The conditions in the
crime lab will also affect the process. Different lighting or different
optical systems will produce different images. Different technicians
will focus their cameras differently. All of this affects image
quality. To make matters worse, many of the images will be nearly
identical. For example, all of the Smith & Wesson 9mm barrels will
produce very similar results. All of the Glock 9mm barrels will look
very similar. All of the Sig-Sauer 9mm barrels will look very similar.
Why? Because large operations like theirs can produce very consistent
products. The differences will be subtle. (By the way: The more subtle
the differences, the more easily they'll be worn away.) The image
processing system will have to make an initial catalog of these guns
that's able to draw some distinction between articles. Obviously, the
problem gets worse as more and more time elapses between the baseline
test and any subsequent test.

Now the data must be stored and easily retrievable. To accurately
record the details in the data, the image files must be large and style="font-weight: bold;">uncompressed. Compression
algorithms, like the JPEG standard, will distort the data. How large
will the record be? I would guess that 2000 pixels by 2000 pixels (4
megapixels) would be sufficient. The images can be gray scale, so that
limits the depth to 8 bits. However, the resulting .BMP files are still
3.81 MB each! And there are three of them per record plus their Fourier
transforms which are also 3.81 MB each!! Thus each firearm in the
system requires at least 15.24 MB of diskspace. File compression can
lower this somewhat, but not by much.

AB35 counts upon many improbable things happening. It assumes that the
guns in the system will not show wear nor will they be deliberately
altered. It assumes that all test labs will produce image files of the
same quality. It assumes the ability to design a database system will
be able to sort through enormous image files to produce a few possible
matches per test subject.

Even if these impossible conditions can be met, there are still more
hurdles to cross. The one that's most likely to get the bill tossed is
especially relevant in this State: How are your going to pay for it?
This system would cost millions to set up, who knows how much to
maintain, and promises very little in return. Maryland has had a
similar system in place for about three years now. They finally solved
a case with it a few weeks ago. Millions that could be used to hire
more police will, instead, be spent on programers. This is good for the
geeks, but not so go for the rest of us.

Wednesday, December 11, 2002

Poor Trenty Lotty. He's up and get something stuck in the wringer.

Truth be told, the Teletubbie from Mississippi didn't say anything worse than what's come out of Grand Dragon Bobby Byrd's (D-KKK) mouth. It's not like the guy was on national television going off about "white niggers". However, it was a pretty stupid thing to say. (Or as the Hooded Wonder from West Virginny might say, "Pretty, pretty, pretty, pretty stupid.")

But should Lott step down?

Yes, but not for this.

Lott has less backbone than a washed up jellyfish. He squandered the GOP's power when he briefly held the Majority Leader's spot in 2001. He has a long and infamous history of appeasing and coddling Democrats. It's fitting that he should lose the top spot in what would be the ultimate example of caving-in to the Dems.

No, he should go because of his stupid remarks; but it make as good an excuse as any.