Anton Roland de Klerk (Antonroland)
Post Number: 8
|Posted on Monday, October 15, 2007 - 04:40 am: ||
So 3rd level detail and its weight / value will obviously then be in line with the quality of the unknown print? It makes sense to me as such anyway.
If I understand you correctly then you also treat absolutely every comparison/conclusion on it's own merits given the print under scrutiny?
Anton Roland de Klerk (Antonroland)
Post Number: 6
|Posted on Tuesday, October 09, 2007 - 01:11 am: ||
Thank you for a great effort there. Much appreciated!
Once I have worked through it all I might throw a question or two back at you purely to be sure I understand your viewpoints correctly.
Post Number: 246
|Posted on Wednesday, October 03, 2007 - 08:10 pm: ||
Short Answer: Show me the prints.
Long Answer: Show me the prints and I can turn on a video recorder and let you know at what point I decide it is an identification. However, let me do the same exam next week and I may begin with a different target set of search features, progress through the examination in a different sequence, and decide that I have an identification at a different point in the exam. Whether you give me small, fragmentary prints or complete and legible palmprints, the point at which I am able to effect an identification can vary. Also, with a large impression, such as a palmprint, I will not compare and use every "point" (level 2 feature) during the examination. If you ask me to compare the same complete and legible palm impressions on different days, I will stop my examination with a decision of sufficiency for identification at different places.
In an attempt to explain the difficult concept of "how much is enough" to identify (individualize), I have often used the analogy of recognizing one's spouse (or partner, parent, sibling, child, etc.) as you walk down the street to where they are standing.
Depending on the quality and quantity of the information available to you due to lighting conditions, distance, geometry of your point of view and other factors, your decision about how much is enough to identify that person (for you to be absolutely certain it is that person) will vary. In bright daylight, when my father is not wearing a hat and I am walking directly toward him, I may be certain it is him by the time I am fifteen feet away. On a foggy night, when my father is wearing a wide brimmed hat and I can only see his face at an angle, I may need to get within five feet before I am certain it is him.
Below is a series of images attempting to partially demonstrate the concept. In each subsequent pair of images, the amount of information and the quality of information is degraded from the previous set.
To create the above pairs, I used the same reduced-size mask, same amount of added noise and same amount of stylized "wind" (in Photoshop) on each pair of subsequent images.
For me, "how much is enough" to individualize my loving wife's face is about Image 2 and the fingerprint is about Image 6.
Below are nine level 2 features marked on Image 6 after it was enhanced with Gaussian blur of 1.5 pixels radius, brightness decreased by a value of 25 and contrast increased by a value of 35 (in Photoshop).
Image 6 Enhanced
I might consider Images 7 or 8 suitable for comparison in casework, but would not individualize either against the Image 1 impression due to the poor quality of the smaller "impression" and the paucity of comparable features (including a questionable bifurcation versus ridge ending in the center area of the impression, and a questionable short ridge or spur versus overlapping ridge endings in the top area). Of course, the open field lacking level 2 features in the top right area of the impression also provides meaningful information for comparison.
The above images could be sampled in an unending variety of combinations, with hundreds of gradients instead of just ten. If sampled with 100 slight variations in the gradient of quality/quantity of information, I believe the answers from Forensic Scientists/Latent Print Examiners would vary insofar as which individual image "is enough" to identify. Different experts would choose different images depending on their personal knowledge, skills and abilities (and biases). Factors such as training and empirical knowledge would cause experts to assign various weights/value to different features and combinations of features. If 100 sample images displayed only slight variation between subsequent (degraded) images, I believe that individual experts would select different images on consecutive days.
Some agencies have internal quality assurance standards governing what cannot be ignored as sufficient for comparison, such as "Failure to claim latent prints having nine individual points (level 2 features) in clear, continuous ridge detail shall, barring extenuating circumstances, result in a memorandum for record generated by the office supervisor for inclusion in the individual expert's personnel file."
Few agencies have internal quality assurance standards requiring an identification if X number of corresponding points are found in common because the amount of "enough" will vary depending on the quantity and quality of information present... also, the interpretation of an "unexplainable discrepancy" is partially dependent on the quality/clarity of information recorded in impressions. Eighteen "matching" level 2 features in an image with poor quality (including pressure distortion and/or photographic factors) may have less meaningful and comparable information than an impression of good clarity, but having only eight level 2 features.
It is never a simple matter of "counting points," as evidenced by the fact that no agency would consider changing the image standard for fingerprints in their AFIS and forensic files to 200 pixels per inch (even though 200 ppi tends to capture almost all level 2 features).
500 pixel per inch representation
200 pixel per inch representation
Experts want to see the ridges, and that means seeing the best quantity and quality of friction ridge skin impression information possible (many agencies are migrating to 1000 ppi images during AFIS upgrades). If experts were only counting points (level 2 detail only), then they would not need higher spatial resolution images. If experts were only counting points, they could effect latent print identifications from skeletonized tracings/representations of friction ridge level 2 detail alone... which does not happen.
Scientific measurement of uniqueness (how often formations of ridges in sequence occur in a given population) is progressing relatively slowly in our discipline (as compared with DNA, for example). However, I believe such scientific measurement of friction ridge skin uniqueness will continue to progress and, after proper validation, be well-embraced and utilized to the benefit of society.
DNA is often lauded as the model which fingerprint uniqueness research should follow, but the subjective determinations made in DNA examinations tend not to be mentioned. For example, "how much is enough" is not a cut and dried answer when DNA mixtures (from multiple donors) are encountered in a stain. With mixtures, it is not uncommon for examiners with differing training, experience and ability to arrive at different conclusions. Also, it is not widely known that DNA examiners may use cumulative fragments of DNA from adjacent stains (not unlike simultaneous fingerprint impressions) to achieve "enough" to identify a person (individualize relative to statistical probabilities and the earth's population).
Until statistical probabilities become a genuine tool in our discipline, I believe we will continue to rely on forensic programs requiring hundreds of thousands of examinations to equip an expert with empirical knowledge of the nature of friction ridge skin impressions and what constitutes sufficient uniqueness to individualize... accompanied by training-to-competency testing and periodic proficiency testing. Some would argue that automated fingerprint identification systems provide a type of statistical modeling in certain circumstances. Computerized comparison of fingerprints has been going on for decades worldwide, yet even the best systems are not yet as accurate at identifying single latent prints as trained-to-competency Forensic Scientists/Latent Print Examiners.
I believe statistical modeling will eventually provide tools enabling individualization of small latent prints that could never be identified using current human expert capabilities (limitations). Also, statistical modeling may enable meaningful correlations of tiny friction ridge skin impressions, such as "corresponds with the person's left thumb and statistically occurs only once in every 28 million persons." Accurate statistical modeling may change the way we effect identifications, strengthening our discipline by confirming sound practices as well as spotlighting errors when experts attempt to stretch beyond statistically valid findings.
The above posting does not purport to represent the position or opinion of the US Department of Defense, US Army, US Army Criminal Investigation Command, SWGFAST or any other organization with which Ed German was or is affiliated. This is just Ed's opinion.
Anton Roland de Klerk (Antonroland)
Post Number: 5
|Posted on Wednesday, October 03, 2007 - 07:51 am: ||
I am not looking for any SWGXXX manual approved correct answer but rather the personal arguments of identification experts.
How and when do you reach the conclusion of positive identification?