Face-Recognition Software Fingers Suspects

eldavojohn writes, "In Holyoke and Northampton, Massachusetts, the police have a new member on the team. It's facial recognition software that will mine the 9.5 million state license images of Massachusetts residents. From the article: 'Police Chief Anthony R. Scott said yesterday he will take advantage of the state's offer to tap into a computer system that can identify suspects through the Registry of Motor Vehicle's Facial Recognition System.' The kicker is that this system been in use since May and has been successful." An article from Iowa a few weeks back mentions that software from the same company (Digimark) is in use to catch potential fraud in applying for driver's licenses in Alabama, Colorado, Kansas, Massachusetts, Oregon, and Texas. But offering the software and photo database as a resource to police departments raises the stakes considerably. I wonder what the false positive rate is.

Comments

finger suspects ?

[Monsieur_F]

$ finger suspect
finger: suspect: no such user.

$ finger suspects
finger: suspects: no such user.

Re:finger suspects ?

[smittyoneeach]

You, for one, have clearly ventured into Soviet Rootkit territory: your new PCI overlord welcomes you!

False positive rate?

[Anonymous Brave Guy]

I wonder what the false positive rate is.

Speaking as someone with (a) some common sense and (b) a formal CS background including image processing work, I think it's fair to say that it won't be zero.

I hope they have good procedures in place to immediately drop any proceedings against those who are misidentified, and that any automatic identification using this system is not somehow considered 100% reliable in court.

Re:False positive rate?

[Adam Zweimiller]

Yeah, but futhermore wouldn't it be safe to say that they don't just go indict someone with charges without a gumshoe comparing the photographs themselves? I mean, theres go to be some sort of human involvement. Lets say the have a CCTV image of a buglary suspect and they use this software to scan the DMV photos for a match, and the software returns 1 or more matches. They don't just throw the match(es) in jail right then. I think its a safe bet that law enforcement would use their own peepers to compare the DMV photographs with the CCTV to see if its close, and then go about questioning the match(es) for their whereabouts..etc..looking for other evidence before going ahead with prosecution. It's obvious that this system is meant to give leads rather than 100% solve cases. Sure there are going to be false positives, it's a computer look for matches. It's more than likely that it's designed to be liberal with its matches simply to give detectives a list of a dozen possible suspects rather than the entire population of a city/town etc. Regardless, I can't say I'm entirely surprised that a slashdot editor took this chance to stir the pot on something that for the most part is cool, useful, and manages to assist law enforcement without trampling our privacy.

Re:False positive rate?

[trianglman]

The issue is more than just the false positive rate. The problem is that they are going through the entire DMV records. As it stands right now, most places can only go through previously arrested people for things like fingerprint and facial matches, which is something that comes with having a record. I, as a law abiding citizen on the other hand, should not be immediately thrown under suspicion just because my face is somewhat similar to a blurry CCTV image, which is what the false positive rate could cause. I have a job that requires me to be in a certain place at a certain time, thats not exactly possible if I am being held for questioning because of something someone I have never met did something on the other side of town. If I could trust our government to use new technologies judiciously and with restraint, it wouldn't be a problem, but this hasn't ever been the case and, short of some utopia suddenly appearing, probably never will.

Re:False positive rate?

[nbauman]

>I wonder what the false positive rate is.

It will be like the Do Not Fly list.

Years ago, Scientific American had a story about a prototype system for facial recognition created of students at Brooklyn College. They had a database of about 1,000 faces, and they showed the 2 most similar and the 2 most different. The 2 most different were very different. The 2 most similar were so similar, I couldn't tell them apart. So back-of-the-envelope, I'd say about 2 faces in 1,000 will be so similar you can't tell them apart.

(Surely on Slashdot somebody must know the current research.)

So if there are 300 million people in the U.S., and you have a common-looking face, you'll have a close match to 300,000 people. Or 8,000 people in New York City.

(In New York a popular Catholic priest was arrested, based on a victim's identification, and charged with rape. His parisioners couln't believe it. Finally the cops found another guy, and charged him with the rape. Finally, they found a *third* guy, and he seemed to be the one. The newspapers published the 3 pictures. They really looked alike. Funny thing was, they were different racial types, too. One guy was hispanic, another guy was Italian.)

Re:False positive rate?

[bob65]

Well getting lies from random people on the streets would be analogous to getting a false positive from this system (except that the system arguably isn't out to frame anyone on purpose), and since police have no trouble dealing with such false positives from "tips", they should have even less trouble dealing with false positives from this system.

Re:False positive rate?

[Duhavid]

I think you are assuming that the police are not having a problem
dealing with false positives from "tips". I suspect that is not
proven.

Re:False positive rate?

[dekkerdreyer]

"Round up twice the usual suspects!" - Louie Renault

Re:False positive rate?

[Entrope]

There is no false positive problem. Police said so on TV!

Explanation: I was flipping through channels and landed on one that was following a Los Angeles SWAT team. They were serving a warrant on an apartment based upon a tip from a confidential informant that there was a gun in the apartment (and I think that drugs were sold out of the apartment -- I missed the start). Nobody answered the door, so the SWAT team battered down the front and back doors, broke a window to investigate a room that was locked, and ransacked the entire place. They found no drugs and exactly one "weapon": a pistol-shaped BB gun. The conclusion from the SWAT team leader: that the confidential informant had proved his worth *and* that their destruction of the apartment had shut down a drug distribution center.

I wish I were making this up. Sadly, this is probably typical of cases that Radley Balko has cataloged.

Oh yeah?

[slughead]

Face-Recognition Software Fingers Suspects

And what does it do if they're male?

False positives before, too

[Wylfing]

I fear "automatic" matching of criminals and trying to catch them, e.g., when they renew their license. Here is a true false-positive story that happened to me. I went to renew my driver's license, and the nice lady informed me that she could not issue me a license because I had had mine revoked in Maryland due to felony charges. Now, I have never committed a felony and I have never been to Maryland, let alone had a driver's license there. The nice lady was unpersuaded by this information. The database said I was a felon in Maryland, and that was the end of the story.

After much yelling about the problem, it was finally revealed that the real felon's name was exactly like mine except for one letter, and some moron doing data entry had gone ahead and decided we were the same person, based solely on name. Since this data problem was local to the "matching" system they had implemented, and not prevalent in who-knows-how-many databases, it was cleared up with a little investigation. However, if that "match" had been replicated into other systems, I could very well have had a nasty time clearing my name. The lady at the DMV was 100% convinced that I was a felon based on what the computer told her. Quite likely, no one else would have believed I was innocent either.

I can see this system playing havoc with people too. I have met people with no connection to each other but who nevertheless look virtually identical.

Re:False positives before, too

[whm]

I can see this system playing havoc with people too. I have met people with no connection to each other but who nevertheless look virtually identical.

This article is a great example of what you've described,

http://nebraska.statepaper.com/pages/drudged/innoc ent.html

In summary: There are two girls that look nearly identical. One of them committed a crime, and the other was put in jail for a week. There are photos in the article.

Good old SQL...

[eyeball]

Part of the SQL better include something like "... WHERE OCCUPATION IS NOT 'politician' " otherwise there's be total anarchy.

but no stats

[troll -1]

Sounds promising for law agencies but given that no caught suspects have been named and that criticism persists that face recognition technology is inherently unreliable, I wonder how much of this is just (sales) hype. I mean, come on, give us some real data where you can say it's effective because ... here are the names of the criminals we caught and it can all be credited to the system.

Re:but no stats

[colmore]

There's a fundamental, mathematical problem for any system that screens large populations looking for a small number of targets.

Let's say your system is 99% reliable, that is to say, 1% of the time it checks a negative it reports a positive and vice versa.

Now you screen 1,000,000 people looking for one suspect, your system turns up 10,001 positives. Which one is it?

This is a problem that has been well-studied in cancer screenings. For certain rare types of cancers, there are nearly 100% reliable tests that nonetheless when they report a positive, are usually wrong.

Now it's fine to say, in the case of the cancer, that the 1% of the population should be informed and then checked via another procedure or something. But when we're talking about a process that fingers potential criminals, and in modern criminal justice where merely being a suspect hurts your life in a myriad of ways (god help you if the information winds up somewhere accessible to google, or worse yet, the case has anything to do with terrorism).

I have the same objection to large-scale wiretapping operations, if anything, the human factor there greatly increases the problem.

I, for one, welcome our new CSI overlords!

[`Sean]

Finally! Inept police departments will be able to solve murders and other heinous crimes using awesome computer graphics in 47 minutes or less...just like on TV!

Enhance...enhance...enhance...

Suddenly...

[Anonymous Coward]

A whole lot of moustaches and beards become the new fasion...

WTF!?~

[sc0p3]

This is exactly like finger printing everyone in the state. Privacy has gone out the window. Making use of photos which people allowed for use on their license, to be used to finger them is criminal.

license photograph archive

[JimBobJoe]

You'd be surprised how many state legislatures never bothered authorizing their respective DMVs to archive the photographs (which is a huge change from the days of the original photo licenses, where only negative was produced and no photograph maintained.)

I just took a look at the MA code and couldn't find anything allowing the photographs to be archived by the registry of motor vehicles. Maybe someone else with a better knoweledge of MA law can find such a law.

This is not an insignificant issue...the archival of the photographs and sharing them to law enforcement, basically without limit and without warrant to access the database, is the practical equivalent of requiring every citizen above the age of 16 to show up at the local police station and be photographed.

I consider the photograph archival of US license pictures to be one of the biggest and least known/understood privacy invasions in the last 10-15 years.

Once a criminal, always a suspect?

[Web Goddess]

I had one criminal conviction when I was 18. It has dogged me my entire life. It is so upsetting to hear people say, oh well, as long as it's only *convicted criminals* who go into these insane database searches.

It's wrong to mass-search drivers' license pictures. It's also wrong to mass-search pictures of anyone who has ever been convicted of a crime. Many, many people have a regrettable misdeed in their past. It's wrong to continue to punish people who have, as was once said, "paid their debt to society."

Penitent and paranoid in California.

Twins

[ms1234]

How does it handle identical twins?