Checkpoints and Borders Policy Debate - Is TSA using ATR as a Front while still viewing nude images?

From the comments to a news article about TSA hiring for PDX (http://www.kptv.com/story/19219049/tsa-hiring-workers-for-portland-international-airport).

A poster named jpminpdx says
When I left I moved back in to the airline industry, that I had been in for a few years prior. It's much more enjoyable. Yes it has stressful moments but I enjoy the job. TSA was awful. I could get into specific details but I would end up writing a book. The turnover rate is extremely high. They're also full of lies. The body scanner? TSA says that nobody views the naked scans. That's BS. There is an office that is manned by a couple of agents who view the naked scans. The scanner itself has a cheesy cartoon style picture that is supposed to be in full view of passengers to make them feel better about going through them. I could go on and on. I know every job has the highs and lows but TSA was always low.

Yeah, I know, it is merely an anonymous post. But has TSA ever officially stated that they have no way of doing this and that they do not do this, eg, view images of passengers while pretending not to. I find it hard to believe TSA would even think of doing this, as it couldn't remain a secret for long, but yet I would not be shocked....

I do not find it hard to believe at all. I won't go so far as to say it's likely, but it's a definite possibility and amongst the list of reasons I do not participate in NoS screening.

They are not smart enough to be devious.

I do not remember the post here, but I recall someone saying they saw an NoS installation in progress and noted there were a few wireless routers among the equipment being installed.

I don't really care about agents looking at my body naked.
I don't imagine that the majority get turned on by naked men.

It seems pretty hard to me to believe that a human could look at the image and generate the yellow boxes in the time it takes for them to come up.

Plausible but unlikely. I think we'd have heard rumblings from more than just an anonymous poster by now if such a widespread conspiracy were in place.

ATR* or not I don't go through the scanners.


* Too many TLAs here. No idea what this stands for but I know it refers to the technology that shows little yellow boxes on stick figures instead of nude images.

Let me check my tin foil hat. OK, its on and working.

I find it plausible that the ATR system is the one used to screen passengers, that there is no communication to the perv booth, AND that the images are still flying by for someone to monitor.

It sort of explains the number of reports that someone got out of the CP without the full screening needed. If they had few bags, got their shoes on fast, they could be on their way before the secondary check occurred.

Stupid way to do a double check screen if true.

Tin foil hat is back off.

I don't really care about agents looking at my body naked.
I don't imagine that the majority get turned on by naked men.

Yeah. I don't care, either--I would have no problem going through the millimeter wave units. The backscatter units are another matter, though--it's not an acceptable use of radiation.

Yeah. I don't care, either--I would have no problem going through the millimeter wave units. The backscatter units are another matter, though--it's not an acceptable use of radiation.

Yes I must sadly say I actually bit-the-bullet and used an MMW a couple weeks ago as I was beyond super late for a flight at IAH (I was last to board as door was closing...). Luckily no anoms and I was done quickly even though my opt outs usually go fast at IAH (not HOU...) I didn't want to chance it as I couldn't take a different flight.

Backscatter I will not do, ever. I even changed flights at El Paso because of an opt out delay. Aside from the reading/posts here I know too many people in the industry that tell me about their design and internal testing findings. It is not pretty.

Yeah. I don't care, either--I would have no problem going through the millimeter wave units. The backscatter units are another matter, though--it's not an acceptable use of radiation.

I don't want these *ssholes looking at my family naked. And particularly when they are lying about it. If it turns out that they are conducting a wholesale deception on the American public, they need to be lined up against a wall.

I don't really care about agents looking at my body naked.
I don't imagine that the majority get turned on by naked men.

Do you care if strangers look at your wife naked?

Do you care if strangers look at your children naked?

Do you care if strangers look at anyone naked without their permission?

Look beyond your own lack of modesty. It's okay by me if you don't care who views your naked body; it's your body and you're entitled to whatever level of modesty you choose to have. But so am I - and my level of modesty is lower than yours. I have the right NOT to be viewed naked by government actors if I so choose, and I should not have to choose between invasion of privacy by AIT or sexual assault by full-body rubdown with genital contact.

Plausible but unlikely. I think we'd have heard rumblings from more than just an anonymous poster by now if such a widespread conspiracy were in place.

ATR* or not I don't go through the scanners.


* Too many TLAs here. No idea what this stands for but I know it refers to the technology that shows little yellow boxes on stick figures instead of nude images.

ATR=Automated Target Recognition. It refers to the software that automatically detects anomalies in the scan data. According to other posters on FT, the scanners equipped with ATR don't even create an actual image any more; the raw scan data from the machine is fed directly to the ATR software for analysis.

I'm sure that it is, but I'm not 100% convinced that the scanner doesn't still create an image that can be viewed in the perv box. Given the number of pat-downs and the flat refusal to re-scan that seems de rigueur for alarms, I believe that those images are rarely viewed, if ever - but they still could be, theoretically.

ATR=Automated Target Recognition. It refers to the software that automatically detects anomalies in the scan data. According to other posters on FT, the scanners equipped with ATR don't even create an actual image any more; the raw scan data from the machine is fed directly to the ATR software for analysis.

I'm sure that it is, but I'm not 100% convinced that the scanner doesn't still create an image that can be viewed in the perv box. Given the number of pat-downs and the flat refusal to re-scan that seems de rigueur for alarms, I believe that those images are rarely viewed, if ever - but they still could be, theoretically.

Ah, thank you! I knew what it referred to but could never translate the abbreviation.

And I agree with your assessment as well.

I'm sure that it is, but I'm not 100% convinced that the scanner doesn't still create an image that can be viewed in the perv box.
Except that, especially in the new installations, there is no such box.

Except that, especially in the new installations, there is no such box.

"the box" need not be local to the scanners. the scan data can be fed to high speed processors for further analysis and flagged for human review if need be.
The question has never definitively been answered as to whether this is occurring. The capability exists.

Except that, especially in the new installations, there is no such box.

"the box" need not be local to the scanners. the scan data can be fed to high speed processors for further analysis and flagged for human review if need be.
The question has never definitively been answered as to whether this is occurring. The capability exists.

Yeah, I think the bottom line on this thread is: "Could it happen?" Yes, it could. There may not be a "perv box" on site, there may not be any motivation to do so, but could, theoretically, the primary mode of defense be ATR, and yet could the capability still remain for there to be the completely R-rated images to be viewed elsewhere as a backup plan? Yes, totally plausible. Likely? No, of couse it's not likely, given the various conspiracy implications (among other issues).


Side note: I chuckle at the use of terms like "high speed processors" on these threads. Really? Is anyone using "slow-speed processors" these days? Come on now. The government has the same access to "high speed processors" as the rest of us do. Money is money.

Side note: I chuckle at the use of terms like "high speed processors" on these threads. Really? Is anyone using "slow-speed processors" these days? Come on now. The government has the same access to "high speed processors" as the rest of us do. Money is money.

Well, maybe you can suppress your chuckle a bit. What I'm referring to is massively parallel processing capable of the high speed throughput that you'd need for this kind of work done on millions of passengers a day. Hadoop and Mapreduce, etc. The federal government certainly does not have this capability. They don't have the infrastructure. A handful of companies have the data center power to pull this off, like Google, Amazon, maybe a couple of others.

Side note: I chuckle at the use of terms like "high speed processors" on these threads. Really? Is anyone using "slow-speed processors" these days? Come on now. The government has the same access to "high speed processors" as the rest of us do. Money is money.

Most of the machinery inspection equipment I use was developed from military applications. The government would only allow it to be sold to industry with very strict limits on the speed at which the equipment operated. The military versions at one time were an order of magnitude faster that the industrial version.

So, no, we do not necessarily have the same access to technology as the government does. Money has nothing to do with it.

"the box" need not be local to the scanners. the scan data can be fed to high speed processors for further analysis and flagged for human review if need be.
The question has never definitively been answered as to whether this is occurring. The capability exists.
One could argue that such should be done in order to allow tuning of the ATR algorithm (the difference being that only a small group of people would ever see the images), but the way the system is being used now, it's quite clear that isn't being done because, in order to be useful, there would have to be a record made, in the machine, of whether something was or was not found for each anomaly and it's easy to see that no such record is being kept.

Well, maybe you can suppress your chuckle a bit. What I'm referring to is massively parallel processing capable of the high speed throughput that you'd need for this kind of work done on millions of passengers a day.
You lost me. The question was "are people viewing images" and you're addressing the question of whether there's some central automated reprocessing of the images.

Well, maybe you can suppress your chuckle a bit. What I'm referring to is massively parallel processing capable of the high speed throughput that you'd need for this kind of work done on millions of passengers a day. Hadoop and Mapreduce, etc. The federal government certainly does not have this capability. They don't have the infrastructure. A handful of companies have the data center power to pull this off, like Google, Amazon, maybe a couple of others.

Given that half the startups in Silicon Valley use MapReduce and Hadoop (you realize Hadoop is just an implementation of MapReduce, right?), and the other technologies you're citing, I'm reasonably certain the government has access to them as well. (Half the startups in Silicon Valley use Amazon's data center, too, though that's beside the point.)

Either way, detecting anomalies on a photo does not require "high speed processors," Hadoop, MapReduce, or any kind of parallel or distributed computing. Detecting anomalies on a photo isn't a parallel computing problem, even at the TSA's scale.

My laptop can do it at TSA's scale.

So yeah, I'll keep chuckling when I hear people talking about various technologies that they don't understand.

You lost me. The question was "are people viewing images" and you're addressing the question of whether there's some central automated reprocessing of the images.

I was pointing out that the viewing activity need not be local either to the checkpoint or to the airport, for that matter, since you mentioned the apparent lack of local viewing booths at the checkpoints. You could have automated processes run regionally that could flag some individuals for further human scrutiny. Given the general incompetence of TSA, the more sophisticated the implementation the less likely any of this is, but the infrastructure exists.
From UshiaHammerfest:
Given that half the startups in Silicon Valley use MapReduce and Hadoop (you realize Hadoop is just an implementation of MapReduce, right?), and the other technologies you're citing, I'm reasonably certain the government has access to them as well. (Half the startups in Silicon Valley use Amazon's data center, too, though that's beside the point.)


I can run Mapreduce on my laptop. Or on a server or on a few servers. That is not the point. As you scale up to handle the kinds of data that would be involved with large numbers of passengers, things choke and run more slowly the more data you try shoving through. And if image processing is involved, beyond the sorting and collating of the discreet passengers, you also have to shard stuff over large numbers of machine cores if you want to hit the speeds that are required. And no, half the startups in Silicon Valley don't have the infrastructure to do this. If they need large scale, they rent it from Google or Amazon or the very small handful of players in this market.

Either way, detecting anomalies on a photo does not require "high speed processors," Hadoop, MapReduce, or any kind of parallel or distributed computing. Detecting anomalies on a photo isn't a parallel computing problem, even at the TSA's scale.

If you are running this stuff on all passengers regionally, no, you cannot do this on your laptop. Image processing can make good use of parallel processing. Halide, for example, developed by MIT, is achieving 70x speed increases by doing this even on smaller scale. Keep chuckling. I'm always amazed by people thinking that the future is going to look just like the past. I never thought the government would be able to achieve near real time decryptions of our financial transactions either at large scale, but apparently, here we are.

"RB sez – “there is a report floating around claiming that TSA still has people viewing the raw Whole Body Imager images even with ATR installed.”

The MMW AIT that operates on the ATR system does not have a remote viewing booth. What the TSOs see, are what the passengers see right there on the back of the machine. It generates a gingerbread cutout image, with highlights over any anomalies found. Hopefully, this will clear up any question you have about ATR and viewing booths.

West
TSA Blog Team"

And no, half the startups in Silicon Valley don't have the infrastructure to do this. If they need large scale, they rent it from Google or Amazon or the very small handful of players in this market.


I didn't say "half the startups have their own infrastructure to do this." I said half the startups in the Valley use MapReduce. Don't misquote me.

And as I specifically said, half the startups in the Valley use Amazon's infrastructure to do it.

If you are running this stuff on all passengers regionally, no, you cannot do this on your laptop... Keep chuckling. I'm always amazed by people thinking that the future is going to look just like the past.


Actually you could, particularly given the ridiculously slow speed at which TSA does it. Since you're so fond of talking about what Google does, surely you know that Google automatically detects nudity on millions of images a day and billions of seconds of video a day. This is *thousands* of (if not more) times the scale of TSA. The amount of machine power need to do this is shockingly small -- it would blow you away how little power is needed.

Reading a bunch of articles on the web doesn't make you a computer scientist. Sorry, but you just don't know what you're talking about here, and it's contrary to the point you made in your OP so I'm not sure why you're trying to make it.

Anyway, I believe the original question was answered.

You could have automated processes run regionally that could flag some individuals for further human scrutiny.
Sure, but why? By then, you have absolutely no information on the actual passenger and what was or was not on their person. What would you accomplish by doing such a thing?

One could argue that such should be done in order to allow tuning of the ATR algorithm (the difference being that only a small group of people would ever see the images), but the way the system is being used now, it's quite clear that isn't being done because, in order to be useful, there would have to be a record made, in the machine, of whether something was or was not found for each anomaly and it's easy to see that no such record is being kept.

A valid point, as this should be standard procedure for continuous quality improvement. Such "double checks" could be used to program out the false positives, where an anomaly shows up but there was not actually a dangerous or prohibited item present.

False negatives (there is no anomaly noted on the gingerbread man but the pax is actually carrying a prohibited item) do not seem to be addressed in any public fashion, though we do occasionally hear reports of test items missed.

"RB sez – “there is a report floating around claiming that TSA still has people viewing the raw Whole Body Imager images even with ATR installed.”

The MMW AIT that operates on the ATR system does not have a remote viewing booth. What the TSOs see, are what the passengers see right there on the back of the machine. It generates a gingerbread cutout image, with highlights over any anomalies found. Hopefully, this will clear up any question you have about ATR and viewing booths.

West
TSA Blog Team"

This really is a hot button with them -- for good reason.

Alas, this "clears up" nothing. It's just another TSA lie. The Gumby display is simply an add-on algorithm for the existing software which produces the voyeur image for the voyeur booth. The machines have to run on the existing software or they don't work at all. So, the clerks can flip a switch and turn off the Gumby display whenever they want. I wouldn't be surprised if they could display the naked image on the new Gumby display because the software is still there.

Okay, my understanding of what ATR does, based on what I've read here on FT, is as follows:

Regular, non-ATR scanner
1) Scan happens
2) Imaging software composes a raster image based on the raw scan data
3) Raster image is displayed on the monitor in the perv booth
4) Raster image is deleted and never stored or transmitted

ATR scanner
1) Scan happens
2) ATR software examines the raw scan data directly to search for anomalies; the raw scan data is not passed to the imaging software at all
3) ATR software creates a gumby image is anomalies are found, or a plain green OK screen if none are found

Now, TSA has repeatedly said, for several years now, that the images are never saved, stored, recorded, or transmitted anywhere but the perv booth.

But they've never said anything about the raw scan data. Is that data stored? Is it cached in the machine's internal hard drives or solid state, non-volatile memory? Can it be used to create an image at any time in the future?

------------------------------------------------------------

This situation reminds me very much of one that occurred at Walt Disney World several years ago.

Disney has, for decades, been fighting not only ticket fraud and counterfeiting, but also has a strange aversion to someone buying a multi-day ticket, using part of it, and allowing someone else use the remainder. They call this "ticket transference", and they hate it so much they were actually able to bribe convince Florida lawmakers to make it illegal to transfer theme park tickets from one person to another.

Part of their strategy was to add biocode ID checks to the tickets. Basically, it's a fingerprint scanner - one scans ones fingerprint when one uses a ticket for the first time, and that scan is used to identify the ticket owner on subsequent uses. If a subsequent fingerprint scan doesn't match the original, the turnstile won't open.

When this was originally put into place, there was a furor over the fact that a huge private company was storing a database of millions of fingerprints. Disney escaped the furor by assuring its guests that the image of the fingerprint would never be stored; only a long, complex numerical representation of the fingerprint would be stored, and used for comparison against subsequent scans. They also claim that all such data associated with any particular ticket is deleted after the ticket is entirely used and becomes invalid.

But what they kept glossing over was the fact that, if the number can be generated from the image, then the image can be generated from the number. So, in essence, they were not storing a completed image of the fingerprints, but they were storing unique, personally-identifiable information about each guest. Whether it's an image or the raw data makes no difference - it's still PII, and it's still stored.

So, when I heard about ATR using raw scan data instead of a rendered image, it made me wonder - has TSA been misleading us all along by telling only part of the truth, as is their habit? Are they really NOT storing or transmitting the images, but maybe they ARE storing raw scan data? Knowing full well, of course, that such data could easily be used at any time to create an image?

It's possible.

But what they kept glossing over was the fact that, if the number can be generated from the image, then the image can be generated from the number.


I haven't read about the Disney case, but that's not necessarily true. If they're using proper 1 way encryption to generate the numeric string, you can go one way but not the other way. This is how passwords work when stored correctly. Only the encrypted form is stored, and when a user enters their password again, what the user entered is passed through the 1 way encryption algorithm and the result is compared against what is stored.

While I don't know what Disney did, if they were doing it properly this is how they would've prevented someone from generating a fingerprint from a string of numbers.

Actually you could, particularly given the ridiculously slow speed at which TSA does it. Since you're so fond of talking about what Google does, surely you know that Google automatically detects nudity on millions of images a day and billions of seconds of video a day. This is *thousands* of (if not more) times the scale of TSA. The amount of machine power need to do this is shockingly small -- it would blow you away how little power is needed.

Reading a bunch of articles on the web doesn't make you a computer scientist. Sorry, but you just don't know what you're talking about here, and it's contrary to the point you made in your OP so I'm not sure why you're trying to make it.

Show me where I claimed to be a computer scientist. I pointed out that the technical platform exists for large scale projects such as this. I will point out that people that have real expertise in certain areas tend to share it (thinking of PTravel et al with legal expertise, or RadioGirl wrt technical aspect of scanners). Since you choose not to shed much useful and detailed information on this, I'm inclined to believe you have real nothing to add. But I could be wrong. If you have something real to say that would educate us, please do so. Without the air of superiority, if you please.

Show me where I claimed to be a computer scientist. I pointed out that the technical platform exists for large scale projects such as this. I will point out that people that have real expertise in certain areas tend to share it (thinking of PTravel et al with legal expertise, or RadioGirl wrt technical aspect of scanners). Since you choose not to shed much useful and detailed information on this, I'm inclined to believe you have real nothing to add. But I could be wrong. If you have something real to say that would educate us, please do so. Without the air of superiority, if you please.

I responded to the air of superiority you presented in kind. If I was mistaken in assuming your "keep chuckling" comment was intended to convey an air of superiority then I recant.

You may never have explicitly claimed that you were a computer scientist, but you did claim to understand issues of a technical nature that it was obvious to me (a computer scientist) that you don't understand. If you want to understand them, by all means ask, and I will gladly explain what I can (and I mean that sincerely with no intended sarcasm or superiority). But as I'm sure you can relate when it comes to your industry or expertise, it can get a little frustrating when people act as an expert in an area in which they have limited or amateur-level knowledge.

That said, I did try to explain above the technical issues behind processing images. Detecting anomalies on a photo (or on a video, for that matter) is not a very compute-intensive task. For example, YouTube sees *billions* of seconds of video uploaded per day. Multiply that by 30 frames per second. Every single frame is scanned for "anomalies" (which in YouTube's case is nudity). Questionable videos are flagged for review by a human. Compute power to do this is significant but not as high as you might think given the amount of data, and the processors in use are no higher speed than your laptop.

Now look at TSA's scale. TSA is looking at two images per passenger. Each passenger is processed in about 10 seconds or so. Multiply that by the number of NOS machines in use at the same time. (How many, maybe 2000 tops?) That means a supposed anomaly-scanning computer needs to analyze ~400 images per second. Even if the number of NOS were an order of magnitude or two higher, that's not a lot of work that has to be done to look for anomalies.

You appeared to say in an above post that it's possible that anomalies were detected on a photo by a central computer somewhere and then flagged for human review (presumably immediately). Could this happen? Absolutely! And there wouldn't need to be a lot of computing power to do it. You certainly don't need any kind of distributed computing to do it at the speed we're talking.

But my point is, there wouldn't be a point in doing this. The machine itself already does that, and a human local to where the anomaly is is already on site to do a more detailed inspection.

...But what they kept glossing over was the fact that, if the number can be generated from the image, then the image can be generated from the number...

This is what concerns me about ATR. They are overlaying the "modest" image on something. If the raw image is not filtered properly, it could be unfiltered at TSA's discretion. I wouldn't be surprised if the raw images were used to verify the accuracy of the "modest" images at some point.

This is what concerns me about ATR. They are overlaying the "modest" image on something. If the raw image is not filtered properly, it could be unfiltered at TSA's discretion. I wouldn't be surprised if the raw images were used to verify the accuracy of the "modest" images at some point.
Let's get the terminology right. By "raw image", one doesn't mean what's been called the "nude image", but rather the actual intensities as seen by the detectors. It needs significant processing to convert that data into either the "nude" or the "modest" image. The question being asked is whether the machine is doing both.

... It needs significant processing to convert that data into either the "nude" or the "modest" image....

That is good to know. Thanks for the clarification.

Let's get the terminology right. By "raw image", one doesn't mean what's been called the "nude image", but rather the actual intensities as seen by the detectors. It needs significant processing to convert that data into either the "nude" or the "modest" image. The question being asked is whether the machine is doing both.

While I'm no expert on image processing, I'm skeptical that the scanner data needs much processing simply to display the data as an image. It probably stores the intensity data as a 2D or 3D array in RAM. That data may not need much more than an actual display buffer to copy itself to. In theory only the computer program is supposed to have access to that array. Consider a 2D array. Call it Intensity[x][y]. So how do you get that Intensity[] array into the display buffer of a remote viewing station? You may just have to copy it to the remote screen where the array can be viewed as an actual image. This connection could be wired or wireless. Of course the program would have to support this RemoteCopy() function. I have no doubt that at least the debug version of the program would be able to display a human readable image to compare against the cartoon one and the alerts. The only question is whether the release version also supports this function and whether the TSA takes advantage of it. Hence the valid concern and paranoia. It would be awfully nice if the TSA could offer evidence that all of the remote perv rooms have been dismantled. They have been caught in lies many times before. Some photos would be a good start to reassure the public that the cartoon images are not just a form of elaborate deception. Another 'layer of security'.

While I'm no expert on image processing, I'm skeptical that the scanner data needs much processing simply to display the data as an image. It probably stores the intensity data as a 2D or 3D array in RAM. That data may not need much more than an actual display buffer to copy itself to.
I'm not an expert in image processing either, but the raw data is in no way like an "image". The intensity data doesn't correspond to pixels in a 2D detector similar to a camera, but rather to intensities collected by a 1D detector moving over an arc. Just displaying those in a display buffer wouldn't produce an image that a human could easily interpret. I'm not familiar enough to know whether multiple pixels of scanner data need to be aggregated to produce a single pixel of image (in somewhat the same manner as a CAT scan, although something very different is going on) or whether it's just a matter of remapping the data from circulat to linear, but, either way, the raw data isn't the image.

I'm not an expert in image processing either, but the raw data is in no way like an "image". The intensity data doesn't correspond to pixels in a 2D detector similar to a camera, but rather to intensities collected by a 1D detector moving over an arc. Just displaying those in a display buffer wouldn't produce an image that a human could easily interpret. I'm not familiar enough to know whether multiple pixels of scanner data need to be aggregated to produce a single pixel of image (in somewhat the same manner as a CAT scan, although something very different is going on) or whether it's just a matter of remapping the data from circulat to linear, but, either way, the raw data isn't the image.

At some point the 2D image will exist for the software to analyze. If it exists it's child's play to copy it.

Last year, I had a chat with an FSD at a particular airport after I opted out. We had a 20 minute interesting discussion about a variety of aspects of security.
He told me that they can see the individual hairs on your arm with their software/body scanners.

The remote viewing rooms have nothing in them anymore after the conversion from AIT to ATR. Also the post about wireless routers is wrong. Wireless is not an accepted form of transmission for the entire reason that it is not secure enough.

Last year, I had a chat with an FSD at a particular airport after I opted out. We had a 20 minute interesting discussion about a variety of aspects of security.
He told me that they can see the individual hairs on your arm with their software/body scanners.

I hope he's lying. That means a lot more rads than they claim.

At some point the 2D image will exist for the software to analyze.
I disagree. Why create that image instead of writing algorithms that operate directly on the raw (unconverted) data? Operating on the raw data will be both faster and more accurate. Although there are some commonalities between the way humans and computers analyze images, there are also fundamental differences.

Wireless is not an accepted form of transmission for the entire reason that it is not secure enough.
I don't see this. Even if you view WPA as not secure enough (and you may), there's no reason why data can't be encrypted before wireless transmission. (Note that I agree there's likely no transmission of the data, but not for this reason.)

I disagree. Why create that image instead of writing algorithms that operate directly on the raw (unconverted) data? Operating on the raw data will be both faster and more accurate. Although there are some commonalities between the way humans and computers analyze images, there are also fundamental differences.

Huh? If you're going to analyze the result for suspicious patterns you need to be able to see patterns in the first place--it's going to be reduced to an array of intensities.

There's no reason to convert that to something like .jpg--but note that when a program displays an image it first converts that .jpg back to an array of intensities. (Albeit with three separate intensities as it's a color image.)

Huh? If you're going to analyze the result for suspicious patterns you need to be able to see patterns in the first place--it's going to be reduced to an array of intensities.
Please re-read what was written. Yes, it's an "array of intensities", but if you were to display those on a monitor, you would not see what you think you'd see because there's a mapping step necessary to reposition and/or merge the pixels to correspond to what looks like an image of the person. That step is needed if a human is analyzing the data, but not if a program is since the program can just as easily (in fact, perhaps easier) work directly on the non-transformed image.

Please re-read what was written. Yes, it's an "array of intensities", but if you were to display those on a monitor, you would not see what you think you'd see because there's a mapping step necessary to reposition and/or merge the pixels to correspond to what looks like an image of the person. That step is needed if a human is analyzing the data, but not if a program is since the program can just as easily (in fact, perhaps easier) work directly on the non-transformed image.

You are quite correct. The image is created for human benefit as it works to our sensing mechanism. The data can be analyzed inside the computer by converting it only to those things it needs to "see." Not having eyes, the computer needs no image to be generated.

Please re-read what was written. Yes, it's an "array of intensities", but if you were to display those on a monitor, you would not see what you think you'd see because there's a mapping step necessary to reposition and/or merge the pixels to correspond to what looks like an image of the person. That step is needed if a human is analyzing the data, but not if a program is since the program can just as easily (in fact, perhaps easier) work directly on the non-transformed image.

What mapping step?

An array of intensities is a black and white image, exactly like we see in the PR stuff they released.

It's possible the intensities are floating point numbers rather than bytes but only other things I see to do are flip it over (Windows works backwards, the lowest address is the bottom of the image) and perhaps align it (I don't recall the alignment rules for black and white images.)

You are quite correct. The image is created for human benefit as it works to our sensing mechanism. The data can be analyzed inside the computer by converting it only to those things it needs to "see." Not having eyes, the computer needs no image to be generated.

But an image in memory *IS* simply an array of intensities!

But an image in memory *IS* simply an array of intensities!

Not exactly, but close. The data to create an image is an array of values. To produce a visible image those values must be read and the pixels that create the image produced. A completely different algorithm can use the same data in an number of different ways, one of which may be to discern areas of anomalies. In these alternate uses of the data, no viewable image need be produced.

Not exactly, but close. The data to create an image is an array of values. To produce a visible image those values must be read and the pixels that create the image produced. A completely different algorithm can use the same data in an number of different ways, one of which may be to discern areas of anomalies. In these alternate uses of the data, no viewable image need be produced.

If the array is in the right format it takes *ONE* line of code to throw it on the screen. I can't think of any reasonable format to hold the data in memory that it couldn't be on the screen with half a dozen lines of code.

To answer the original question: no. That's too big a conspiracy for a public agancy like TSA to manage.

But, yes, the machines are still capable of transmitting human readable images. 1. That would be the only way to 'verify' they might work. 2. The managers and developers would be too scared to remove that code for fear of breaking something.

As for the images, I'll bet a whole dollar that the sensors emit some form of RAW image, maybe even TIFF based. Why? Well, the hardware designers have no particular interest in how the images will be used but they want their hardware used as much as possible. They also don't want to commit to a custom format because that would expose them to a never ending stream of 'enhancements.' Thus, they just provide a proven, familiar format.

This is also good for the ATR developers since there's scads of libraries for converting, manipulating and analyzing standard image formats.

There's nothing remotely special enough about ATR that would required a completely custom format, except that it's government funded so all bets are off.

As for the images, I'll bet a whole dollar that the sensors emit some form of RAW image, maybe even TIFF based. Why? Well, the hardware designers have no particular interest in how the images will be used but they want their hardware used as much as possible. They also don't want to commit to a custom format because that would expose them to a never ending stream of 'enhancements.' Thus, they just provide a proven, familiar format.

Agreed. There's no reason to reinvent the wheel. It's a standard image format.

I'm not an expert in image processing either, but the raw data is in no way like an "image". The intensity data doesn't correspond to pixels in a 2D detector similar to a camera, but rather to intensities collected by a 1D detector moving over an arc. Just displaying those in a display buffer wouldn't produce an image that a human could easily interpret. I'm not familiar enough to know whether multiple pixels of scanner data need to be aggregated to produce a single pixel of image (in somewhat the same manner as a CAT scan, although something very different is going on) or whether it's just a matter of remapping the data from circular to linear, but, either way, the raw data isn't the image.

Wow - normally I agree with most of your posts, but I think you are truly missing the mark here.

1) The only difference between the 2-D image that a digital camera captures and the 2-D image of an XRD is the time it takes to capture the data. Both machines record the intensities of various wavelengths of light, pixel-by-pixel, and store the data in a common image format. Both output files can be rendered human-visible using a simple decoding algorithm. No human could recognize the raw data from a digital camera image, either.

A black-and-white 1 MP camera has 1 million detectors that fire almost simultaneously, and the camera records the radiation intensity sequentially from each detector. The X-ray has one detector that it moves to a million different places and records the radiation intensity sequentially from each position. The end product is identical.

2) Your concept that there is an elaborate process required to convert raw data into an image is mistaken. Each 'pixel' is simply the intensity of the detected beam in a single spot. So in a simplified example, you have something like this:

00000000000001000000012100000122210001111111000000 0000

This is a string of raw data captured by the camera CCD or the X-ray detector. The only thing that is required to turn this data into an image is to add a header and footer that indicate the general size of the image, the length of each row, and some other basic information.

[jdi:54,9,6]00000000000001000000012100000122210001111111000000 0000[]

Now this raw data has become an image file. The [jdi: 54,9,6] indicates this is a JanetDoeImage, size=54 pixels divided into 9 rows and 6 columns. So a simple command to render the data into a 9x6 array gives you

000000000
000010000
000121000
001222100
011111110
000000000

For .jdi files, the monitor displays a bright green light where there is a "2", a dim green dot where there is a "1" and a black dot where there is a "0". Voila, a triangle.

While this is a minor simplification over common image files like .tif or .jpg, the basic fact is that any string of raw data can be easily and trivially converted into an image, especially for black-and-white images like X-rays and MMW. For any image format, the data is stored in a simple linear array, with a header/footer that explains the logic of where to break up the array. You could download free software that recovers corrupted images, feed it the raw data string, and it would be able to generate a very reasonable facsimile of the original image, even without the header and footer information.

So while there is clearly a definitional difference between "raw data" and "image file" and "image", in practice the ability to convert raw data into image is such a trivial exercise that the the distinction is meaningless. If I take a photo of someone in a .jpg file and convert the ones and zeros into a text file, and e-mail it to a third party, no human would be able to recognize it as an image immediately. But it would take a good programmer 2 minutes to construct an image from the text file, and it would be hard to argue with a straight face that I hadn't sent an image.

3) There is no doubt in my mind that all the original software capabilities for image viewing and storage exist in the current ATR software. All they did was add a processing step (edge detection) on top of the old software. Whether any regular front-line TSA employee has access to the images rendered on a computer screen is a separate question, and I do not know the answer.

4) Yes, the raw data is not the image. Therefore, it is quite possible for TSA to deny storing "images" while still storing all the raw data necessary to construct the images with a simple command or algorithm.

I do find it hard to believe that Facebook can implement facial recognition (http://www.pcworld.com/article/229742/why_facebooks_facial_recognition_is_creepy.html), but the NSA/CIA/FBI/DHS/TSA is not taking advantage of the wealth of personal images taken by the TSA every day. If I was tasked with the job of identifying and stopping terrorist attacks on airplanes, I would transmit the data for every face that walks through the AIT to a national database that uses facial recognition software to check against photos of known or suspected terrorists. Hits or potential hits could be referred to agents for "beverage testing" or "checking IDs at the gate." :D

Wow - normally I agree with most of your posts, but I think you are truly missing the mark here.

1) The only difference between the 2-D image that a digital camera captures and the 2-D image of an XRD is the time it takes to capture the data. Both machines record the intensities of various wavelengths of light, pixel-by-pixel, and store the data in a common image format. Both output files can be rendered human-visible using a simple decoding algorithm. No human could recognize the raw data from a digital camera image, either.

A black-and-white 1 MP camera has 1 million detectors that fire almost simultaneously, and the camera records the radiation intensity sequentially from each detector. The X-ray has one detector that it moves to a million different places and records the radiation intensity sequentially from each position. The end product is identical.

2) Your concept that there is an elaborate process required to convert raw data into an image is mistaken. Each 'pixel' is simply the intensity of the detected beam in a single spot. So in a simplified example, you have something like this:

00000000000001000000012100000122210001111111000000 0000

This is a string of raw data captured by the camera CCD or the X-ray detector. The only thing that is required to turn this data into an image is to add a header and footer that indicate the general size of the image, the length of each row, and some other basic information.

[jdi:54,9,6]00000000000001000000012100000122210001111111000000 0000[]

Now this raw data has become an image file. The [jdi: 54,9,6] indicates this is a JanetDoeImage, size=54 pixels divided into 9 rows and 6 columns. So a simple command to render the data into a 9x6 array gives you

000000000
000010000
000121000
001222100
011111110
000000000

For .jdi files, the monitor displays a bright green light where there is a "2", a dim green dot where there is a "1" and a black dot where there is a "0". Voila, a triangle.

While this is a minor simplification over common image files like .tif or .jpg, the basic fact is that any string of raw data can be easily and trivially converted into an image, especially for black-and-white images like X-rays and MMW. For any image format, the data is stored in a simple linear array, with a header/footer that explains the logic of where to break up the array. You could download free software that recovers corrupted images, feed it the raw data string, and it would be able to generate a very reasonable facsimile of the original image, even without the header and footer information.

So while there is clearly a definitional difference between "raw data" and "image file" and "image", in practice the ability to convert raw data into image is such a trivial exercise that the the distinction is meaningless. If I take a photo of someone in a .jpg file and convert the ones and zeros into a text file, and e-mail it to a third party, no human would be able to recognize it as an image immediately. But it would take a good programmer 2 minutes to construct an image from the text file, and it would be hard to argue with a straight face that I hadn't sent an image.

3) There is no doubt in my mind that all the original software capabilities for image viewing and storage exist in the current ATR software. All they did was add a processing step (edge detection) on top of the old software. Whether any regular front-line TSA employee has access to the images rendered on a computer screen is a separate question, and I do not know the answer.

4) Yes, the raw data is not the image. Therefore, it is quite possible for TSA to deny storing "images" while still storing all the raw data necessary to construct the images with a simple command or algorithm.

I do find it hard to believe that Facebook can implement facial recognition (http://www.pcworld.com/article/229742/why_facebooks_facial_recognition_is_creepy.html), but the NSA/CIA/FBI/DHS/TSA is not taking advantage of the wealth of personal images taken by the TSA every day. If I was tasked with the job of identifying and stopping terrorist attacks on airplanes, I would transmit the data for every face that walks through the AIT to a national database that uses facial recognition software to check against photos of known or suspected terrorists. Hits or potential hits could be referred to agents for "beverage testing" or "checking IDs at the gate." :D

+1

Here's a report about the CBP providing data about license plate scans, done at the border, to a private insurance organization.

http://arstechnica.com/tech-policy/2012/08/license-plates-scanned-at-border-data-shared-with-car-insurance-group/

This wasn't something the DHS/CPB openly admitted to the American public, or requested permission to do. No, this was only determined as a result of a FOIA request by EPIC.

The parent government agency of the CBP, the DHS, is currently at the center of multiple lawsuits and even Congressional investigations(e.g. Fast and Furious) regarding improper and illegal activity. Even today, DHS agents have filed a lawsuit regarding their orders to break the law regarding immigration.

Now, there are members of this board that feel there is little or no need to doubt the veracity of the TSA regarding what data they use, how they keep it or even the safety and propriety of the manner used to get the data(e.g. improper gropes and nude pictures with xray style devices). This is suggested to us of an agency that is known to lie, obscure the truth and facts and even creates secret rules that it expects citizens to follow, without question, even as random agents randomly changes the rules that are posted by the TSA stating those don't change.

The TSA, and the DHS, have done nothing to earn my trust or give me even a small nugget of faith in their ability to do their jobs or follow the laws of this country. Every thing stated by that agency I view with extreme skepticism.