Generic Conditionals

In my retirement from police service, I'm busier than ever. One of the projects that I'm involved with is the creation of an instructional program in report writing for a national police service. In defining the instructional problem, I've found that the learner population has a problem with "factual conditionals." I've also noted this problem in the report writing samples of their forensic science practitioners.

Because people have problems with the relationship between the dependent and independent clauses, their reports are hard to read and interpret. What should be a clear statement - (dependent variable / action) resulted in (independent variable / result) - is often a confusion of meanings.

Often, what should be written as a conditional is written as a declarative statement. This problem hides potential meanings, and obscures avenues for inquiry.

For example, one of the sample videos that I use in my Content Analysis class (link) examines a traffic collision scene. A collision occurs as V1 attempts to turn left whilst exiting a parking garage. In the declarative statement, fault is obvious - turning left eludes to issues of right of way. What is missing is the conditional. If V1's progress is purposefully impeded, then the inquiry turns from a simple traffic collision to a "staged collision," - an entirely different line of inquiry.

When the responding officer records the statements of those involved, as well as witnesses, it becomes important to consider the statements in a "conditional sense." If Person 1's statement is true, then the scene would be arrayed thus." or "If Person 2's statement is true, then Person 1's statement is untrue." The conditional statements help frame the analysis of the statements and the evidence.

Using an example from the weekend's posts, "If headlight spread pattern analysis is a subset of digital / multimedia forensic science (comparisons), then the analysis must examine the recording of the pattern and not the pattern itself."

Just something to ponder on this beautiful Tuesday morning. Have a great day, my friends.

Fusion-based forensic science

When I first started out in multimedia forensics, there were just a few vendors offering tools to practitioners. Ten years later, there were a whole bunch. A further ten years later, and there are a relatively small group of vendors again.

When I sat down with Dr. Lenny Rudin of Cognitech a few weeks ago, to catch up and see what's new with him, we took that walk down memory lane that is customary for the "old guys." We talked all things Cognitech, and the state of the market.

I asked him about "codec / format support" in his tools and it seems that he favors the approach taken by the folks at Input-ACE - or the other way around I guess, since Cognitech has been around much, much longer than Input-Ace. The approach, seize the device and process the hard drive directly, is where Input-Ace is going in their partnership with Cellebrite and DME Forensics. Cognitech is there as well, seemingly with SalvationData.

This fusion-based approach makes sense. Do what you do, well, and let others do their thing well - then partner with them. It makes sense.

On the other end of the spectrum is Amped SRL. I remember standing in the exhibit space at LEVA Asheville when a certain developer was trying to offer his services (even his IP) to Amped. The response he received was that everything that is necessary to the task should be in FIVE. No plug-ins. No extra programs. That, and that Amped was happy with their current staff and development pace. In that conversation, Amped noted that FIVE should be the "go-to" for analysts. For a few years, this was the case. Now, increasingly, it's not.

Is everything that is necessary for an analyst's work in FIVE? Currently, there are features available in DVRConv that aren't available in FIVE. The main one that most customers wanted (when I was fielding calls in the US for Amped Software, Inc.) is the ability to "convert" everything. You see, in DVRConv, you just drag/drop a bunch of files and folders into the interface and it "converts" everything that's convertible - and maintains folder structure. FIVE, on the other hand, only "converts" one file type at a time, one folder at a time. The new "change frame rate on conversion" functionality is another example of something that is done in DVRConv that isn't done in FIVE.

Then, there's Reply. It seems that rather than continue to develop FIVE, Amped is cannibalizing it. Is Redaction moving over to Replay? It seems so. Yes, Replay now has improved tracking functions. But, what about all of those customers who bought FIVE for redaction. How are they feeling right now knowing that promises made about future functionality weren't promises kept. Promises made that weren't kept, you ask? Yes. Amped has yet to make good on their promise, made at LEVA Scottsdale, of adding audio redaction functionality to FIVE. Perhaps audio redaction will make it into Replay. Perhaps not.

Amped was once known as the provider of "specialists'" tools. Now, it's "generalizing" them as it migrates FIVE's functionality to Replay. A lot of US based agencies bought five years worth of support and upgrades when they purchased their original licenses. I'm wondering how that ROI looks now, given Amped's latest moves.

Meanwhile, Amped's competition has closed the gap and begun to overtake them in the marketplace. Dr. Rudin, sensing this, is about to make a big push with some new tools (hint: video authentication).

But, with all of this in mind, probably the oddest development is Foclar's entry into the US market. Foclar's Impress isn't a new product. It's been available in Europe for years. But, in terms of development and functionality, it's about where Amped was on FIVE's entry to the US market ten years ago. Let's see if Foclar has the wherewithal to play catch-up. As Amped raises it's prices for its products and services beyond what the market will bear, they'll leave the door open for Foclar to make some headway - should that European company choose to compete on price.

Gone - for the most part - are Signalscape, Salient Stills, and Ocean Systems. These companies still exist as fractions of their former selves. Signalscape's StarWitness is gone - replaced by interview room recorders and evidence processing machines. Salient Stills never really owned the IP in VideoFocus - which is now a part of DAC / Salient Sciences. As for Ocean Systems, it seems as though it exists at the will and pleasure of Larry Compton (who might be it's only employee). They're still offering the Omnivore, the Field Kit (Omnivore + a laptop), and Larry's training courses. Legacy products are on autopilot. Has ClearID been substantially improved / updated since Chris Russ' sacking years ago? I don't think so. I never understood the value proposition of taking 8 of the algorithms from OptiPix / FoveaPro (which were collectively priced at about $500 for about 70 algorithms) and charging much, much more. Sure, you got a new UI. But was that worth the price? I still have an old MacBook running PS CS2 just to run Chris' old stuff.

Have a great day my friends.

Review: Forensic science. The importance of identity in theory and practice

A paper was recently published over on Science Direct (link) exploring what may be a "crisis" in forensic science. Here's the abstract of the paper:

"There is growing consensus that there is a crisis in forensic science at the global scale. Whilst restricted resources are clearly part of the root causes of the crisis, a contested identity of forensic science is also a significant factor. A consensus is needed on the identity of forensic science that encompasses what forensic science ‘is’, and critically, what it is ‘for’. A consistent and cogent identity that is developed collaboratively and accepted across the entire justice system is critical for establishing the different attributes of the crisis and being able to articulate effective solutions. The degree to which forensic science is considered to be a coherent, interdisciplinary yet unified discipline will determine how forensic science develops, the challenges it is able to address, and how successful it will be in overcoming the current crisis."

The article seems an exploration of struggle for identity, as the title suggests. What is it? What are we to do with it? What happens when it's not done correctly? Who's responsible for reform? Curiously, even as the paper notes the work done on forensic science in the US, it omits reference to the 2017 A Framework for Harmonizing Forensic Science Practices and Digital/Multimedia Evidence (link). I point this out because the OSAC's document provides a solid definition of forensic science that can serve as a foundation from which to explore the paper's topic, as well as to provide a path forward for research and practice.

As a reminder, the definition of forensic science, "The systematic and coherent study of traces to address questions of authentication, identification, classification, reconstruction, and evaluation for a legal context." "A trace is any modification, subsequently observable, resulting from an event."

I think that within the OSAC's definitions, the goals outlined in the paper can be achieved.

Nevertheless, the likely root of the "crisis" as observed by the author can be seen in another area altogether - bias.

The topic of bias, and the many places it influences the justice process has been explored in depth. Particularly, these two papers (link) (link) explore the impact of bias from the standpoint of each of the stakeholders and the influence of modern media.

Another place where bias occurs is in the selection and continuation of cases by the prosecution. For example, what effect does linking a prosecutor's "win / loss" record to their promotability within their organization have on their decision making process? Once they've filed a case, is there evidence of an "escalation of commitment" when confronted with problems in proving their case? With "escalation of commitment" bias, the prosecutor may seek out "fringe techniques" in an attempt to support their theories of the case. Is it these "fringe techniques," not forensic science, that have contributed to the observed "crisis" - as was illustrated in Netflix's Exhibit A?

Still, I'm glad to see that people are starting to identify that there might be a problem in the practice of "forensic science." How that problem, or "crisis," is addressed will make all the difference in the world.

Have a good weekend my friends.

Can Amped's Authenticate assist in detecting "deep fakes?"

Short answer: no.

Long answer: everything old is indeed new again.

In 2016, I presented an information session on authentication at the LEVA Conference in Scottsdale, Az. Here's the session description:

Understanding Concepts of Image Authentication Workshop (2016)
This workshop is for those interested in the authentication of digital images. The workshop provides an overview of the techniques and skills necessary to perform basic authentication examinations using Amped Authenticate (Axon Detect) on digital images in a “forensic science” setting as well as to package, deliver, and present those findings in their local court room context.

At this year's LEVA Conference, it seems that the old topic is being dusted off, with one small addition. Can you spot the main difference between the 2016 session description (above) and this year's (below)?

Authenticate: The Beginners Guide to Image Authentication* (2019)
Image authentication techniques have multiplied over recent years. The simplicity of Image editing and the increase of bogus imagery, “Deep Fakes”, being identified in the media has, quite rightly, meant that methods to detect manipulation must be available to the legal system.

The difference: "Deep Fakes"

There's one big problem though. A "deep fake" is a video. Authenticate doesn't work on video, only images of a specific file type.

In my forensic multimedia analysis course (link), I feature a number of proposed techniques that address so-called "deep fake videos." All of the solutions work within a fusion-based methodology - requiring different tools and applications of those tools to identify the many components that make up fake videos.

But the wider question, why use the term "deep fakes" incorrectly to market an information session on an image authentication tool? Is this SEO gone wild? I don't know. At a minimum, it's confusing. At the maximum, it's deceptive. Hopefully, you weren't just going to learn about "deep fakes." If so, you're sure to be disappointed - again, Authenticate does not process or authenticate video.

Have a great day my friends.

keywords: audio forensics, video forensics, image forensics, audio analysis, video analysis, image analysis, forensic video, forensic audio, forensic image, digital forensics, forensic science, amped five, axon five, training, forensic audio analysis, forensic video analysis, forensic image analysis, amped software, amped software training, amped five training, axon five training, amped authenticate, amped authenticate training

The best artists steal

It's an old adage in the art world that good artists copy the work of their masters, but the great artists steal it (link). With that in mind, a reader of the blog pointed out the apparent similarities between the Camera Match Overlay in Input-Ace (link) and a patent on file at the USTPO (link).

If you read the marketing around the Overlay tool, you might begin to notice the similarities between it and the patent. But, don't be fooled. US Patent 9,305,401 (link) describes a system for building out a scene in 3D utilizing images from that scene, then conducting various measurements. Overlay does none of that. Overlay does exactly what the name implies - it allows you to take the user interface of Input-Ace and "overlay" on top of the user interface of the tool with which you are actually conducting your measurement exam. The idea is that you can infer the location of images present in the images loaded in ACE, and overlaid on the measurement tool, from the data in the measurement tool. Because of this, I don't think ACE is infringing on anyone's patents - in my opinion.

Yes, the patented process is operationalized into Cognitech's software, and that software is actually reconstructing the scene and performing the measurement exams. No, Input-Ace's Camera Match Overlay tool is not reconstructing a scene and is not used directly in conducting the measurement.  Yes, it does lend a hand in attempting to calculate the range of potential measurement values (and thus the error potential in the measurement), so you'll need to be extra careful in how you report and present your results using their methods. You'll also need to explain how you validated your unique results. Yes, you should validate any tool used in your work as well as the results that lead to an opinion / conclusion.

Thanks for reading. Keep the comments coming. Have a good day my friends.

Detecting 'Doctored' Images

Just a wee reminder that authenticating multimedia evidence (aka 'spotting doctored images on the internet') is a bit more complicated than finding the "doctor" in the image.

If you'd like to learn the underlying science behind authenticating this complicated evidence type, check our our course - Forensic Multimedia Authentication (link). Offered on-line as micro-learning, seats are always available and you learn at your own pace. Click on the link for more information, or to sign up today.

Everything old is new again

Last month, I took a look at the new Camera Match Overlay feature in Input-ACE. The Overlay feature can be used in conjunction with a 3D laser scanner and it's accompanying software to create demonstrative exhibits.

I'm not a big fan of the feature, preferring single image photogrammetry to conduct measurements within the evidence items without creating brand new exhibits. But, it occurred to me ... one of the nice things about getting old is you get to see history repeat itself.

When I first arrived at the LAPD in 2001, it had a video / image processing workstation from Cognitech. It was then that I first met Dr. Lenny Rudin. Last month, I was in Pasadena to present a lecture and ventured over to CalTech's amazing restaurant to have lunch with Dr. Rudin and catch up on what's new with him and Cognitech.

Our conversation bounced all over the place and eventually exceeded the allotted time that the restaurant had for lunch service. I like those conversations where time ceases to be a factor, just enjoying the topics and the company.

Of course the current state of the industry came up. Who's who and what's what. If you're reading this and you don't know the names (Cognitech and Dr. Rudin), that's a shame. Dr. Rudin is one of the founders of this thing we now call Forensic Multimedia Analysis, but has largely been written out of the history by those with a more commercial agenda. It's only the old folks who know the likes of Dr. Rudin, Dr. Russ, and the like.

Nevertheless, we discussed input-Ace's Overlay feature a bit, noting it's similarity to Cognitech's Measure package that began life in the 90's (now called AutoMeasure).

Everything old is new again. You can find the 1995 paper that describes Cognitech's Measure over at the SPIE (link). It's not an "overlay" procedure as such, but a single image photogrammetry method that builds out the 3D space within the 2D image. Unlike Overlay and its mixed-methods approach, Cognitech's Measure does have a validation history available in the literature.

Over lunch, we talked about where Cognitech's tools are now and what are the plans for the future of Cognitech's offerings. It has been a while since I've used Cognitech's tools. I'm looking forward to getting to know the new versions and the new products. Hopefully, if Dr. Rudin agrees to allow it, I'll be showcasing some of them in future blog posts.

Have a good week my friends.