The holistic approach

[Dan Perkins]

Hi,

does anyone see a conflict in using the terms 1st, 2nd and 3rd level detail in the context of a holistic approach?

In 2005 the now defunct the Thrid Level Working Group of the now defunct UK National Fingerprint Board published the following reccommendation:

“The Third Level Detail Working Group acknowledges that within the
international fingerprint community, the terminology, ‘1st, 2nd and 3rd
Level Detail’ is used. This group recommends that because of the holistic nature of fingerprint identification, there is no need for the use of these terms within the comparison process. ALL information assisting with establishing the identification of an area of friction ridge detail will be termed, ‘features’.

I would appreciate any thoughts you may have.

thanks

Dan Perkins

[Gerald Clough]

For the sake of argument, I see some differences in what's known about the physical formation of three three levels of feature. ("Level" might not even be a very good word.) We know a good bit about the frequency of L1 patterns, as we define those patterns, in the population, even though the definition criteria are rather arbitrary and have no real import other than taxonomy. We know something of the frequency of L2 detail types, again as we define them, but not much else about frequencies of particular arrangements. We know much less about pore unit configurations, although we might know all that can reasonably be known. We have a long historical perspective on the use of L2 for identification. We also have some historical perspective on L1 as having some identifying power when we have multiple fingers. We have some historical and some very compelling examples of L3 that I think most examiners would accept as fully identifying. So, they are different, and they all have different powers for a variety of different reasons. And we tend to gain knowledge relating to each level type in separate research operations, and inquiries into each of them are at different stages of maturity. There's probably not much more being learned about L1. Since L2 is the grossest level of feature with real identifying power, most of the studies of various examination issues revolve around L2. L1 studies tend to look strictly at L1.

But as to using the Ln terms, there's a good reason for terminology so we can talk about things in a less confusing way than "these small things," "these rows of those small things." "this point where those lines diverge," and "see how these rows tend to run this way." But I do use "features" a lot in writing about examination, exactly to point up that it's not about one thing - it's about ridges, ridge units, how they formed, and how various points where they behave in certain ways relate to other point where they behave in certain ways. But taxonomy is a pretty fundamental part of scientific and technical knowledge and discourse. And discriminating power is generally demonstrated by finer degrees of classification, not more general. And discrimination is what we do. And since we are in the expert opinion business, and the crux of an argument may be centered on arguing the presumed powers of different levels of detail, I think they can't be abandoned. But using the terms is in no way in conflict with a holistic examination. I would go so far as to say you can't do a proper examination except holistically. It seems to me that would imply you were prepared to ignore any but one class of discriminator.

[Neville]

Hi Gerald
'I would go so far as to say you can't do a proper examination except holistically. It seems to me that would imply you were prepared to ignore any but one class of discriminator.'

I am not so sure Gerald that I can totally agree. With the advent of AFIS we no longer need to worry about 1st level if we are search partials with no pattern visible and often are comparing against 'inked' impressions (live scan included) that show no 3rd level features yet we make totally secure identifications on a daily basis. We can conclude that a print is not the suspects on only 1st level detail without even thinking of looking at 3rd level.

Otherwise I like what you have stated, keep up the good work. Your submissions maybe verbose but they always make me think.

[Ernie Hamm]

Aspects concerning the holistic use of “features” were addressed in my presentation, “Friction Ridge Levels of Detail: Bane or Boon?”, to a joint session of the Florida/Georgia IAI in 2005 and The Fingerprint Society at Tulliallan, Scotland in 2006. The FP Society version was published in Fingerprint Whorld, 32:125, 2006 (but “Boon” came out as “Boom”??).

[Dan Perkins]

Gerald, Neville, Ernie many thanks for your thoughts.

Dan

[RL Tavernaro]

... With the advent of AFIS we no longer need to worry about 1st level if we are search partials with no pattern visible and often are comparing against 'inked' impressions (live scan included) that show no 3rd level features yet we make totally secure identifications on a daily basis. ....

Although I am generally in agreement, I can't quite totally agree as you refer to not worrying about 1st level detail in some instances with the use of AFIS.

If we understand ridge flow (even in the absence of classifiable pattern) to also be level 1, I believe some AFIS search algorithms include consideration of the ridge flow as related to recognized level 2 features.

In addition, when it comes time to actually do the comparison, I don't see how we can escape including observation of the ridge flow in the two impressions. Would you ever consider making an identification (individualization) using only the level 2 feature map overlay, without seeing the underlying ridge structure?
Regards, RLT

[Gerald Clough]

Neville,

What I meant was that it's not appropriate to go into an examination determined to use one part of available data and ignore any other available data that has some power, because that's a poor analytical mindset. But it's true that when L1 is clearly not in agreement, we'd be unlikely to do a close analysis of L2, searching for agreement. So it's not such a big deal. And L3 is, to me, more problematic. L3 analysis is not nearly so mature, or maybe it's just that the nature of L3 precludes the kind of confident characterizations we assign to L2 we observe. Either way, L3 is not on very firm ground. For instance, if we propose to individualize from L3, would we not also have to be prepared to exclude from L3? If we could clearly discriminate in a very confident way, as we can with L2, it would have the same power to individualize or exclude. How grossly would L3D in one impression have to diverge from L3D in the other to quash an individualization from L2D that was otherwise sufficiently in agreement? I think it's so difficult to say or even present examples without leaving a very large number of undecidable examples that L3 is largely out of play, and current practice reflects that. Would any examiner exclude an otherwise well-founded identification on account of dramatic apparent differences in L3? I would have to say that there could be L3D observations that would at least pull the examiner up short and prompt some hard thought and/or consultation. But I think it would have to be pretty obvious, and as we significantly note when we see that L2 identifications within the standard or practice have never been shown to misidentify, nothing so obvious has been observed and presented to suggest L2 identifications are unreliable.

So I'm left kind of undecided, but I think I could express it. Is a rigorous analysis of available L3 required in an examination? No, I wouldn't say so. The confidence in fingerprint identification refers to L2D identifications. So in that sense, I can't hold to the strict statement that it can't be properly done unless holistically, since that would imply the same rigor with L3 as with L2, and the reliability of L2 identifications as practiced is well-established. Can L3D be completely ignored when available? No, I wouldn't say that, either, because I'd then be saying that even the most clear gross difference in L3D should be ignored. But there's no reason at this time to believe an identification strictly on L2 as the only available observation is an any less confident opinion than it would have been had L3D been observable. In other words, technologies that don't provide clear L3D have not been shown to weaken identification conclusions from their recorded images. But we remain holistic, in that we do make appropriate use of all available data, and we are always able to say we observed all available data and can explain any weight given to each, whether it be considerable or none at all.

(I'm not precluding identification from L3D. It can be a perfectly viable and appropriate expert conclusion. It's just that it doesn't have the benefit of having been a common practice and the vast number of opportunities to have been shown to be unreliable that conventional L2D identifications have had. It's therefore subject to more vigorous challenge, but it's no less a valid opinion.)