The common assumption of the court, attorneys, jurors and many latent examiners is; AFIS systems are close to 100% accurate, so one search should suffice. Most systems today will have an accuracy less than 75 t0 80 % for latent searches. Typically the vendor will produce an accuracy 75 to 80 % if the latent image has 15 points and the target has the same 15 points. The other thing to remember about these tests are the vendor is using a small database that may not accurately reflect the a larger database of the live system. What can add to the confusion is the tenprint to tenprint accuracy which for the FBI IAFIS is 98% and most state systems probably 95%.
The other factor to remember is the system will produce a number of very good suspects which can move the suspect down the list or when dealing with poor quality images an examiner may miss the identification.
Another factor to consider, minutiae placement which is critical to any AFIS search. Minutiae can be selected by the system and manually edited by the examiner or manually selected by the examiner. Anecdotal evidence would suggest manually placing the minutiae will produce a more accurate result. Studies have shown that an examiner may vary slightly on minutiae selection the 2nd. time they see a latent print (table 3).
The results of any of the above could lead to a wrongful conviction and individual not being interviewed who may help in the investigation or an element of doubt in a jurors mind. To insure a proper AFIS search the prosecution should insure that latents that may impact a case if unidentified are searched more than one (1) time when they have their meeting prior to the court date. The defense attorney on the other hand would be the fail safe, insuring that the latent was searched multiple times.
Bob McAuley Dir. Operations/Training Forensic Biometric Identification Solutions LLC.
Posts
5 comments:
You left out a few important concepts.
1) AFIS is a garbage in/garbage out system. This means that the quality of the latent in the database can directly influence the results of the search as well. The most accurately coded latent will fail to produce the 'match' in a candidate list if the reference image is poor quality.
2) The goal of a latent examiner using AFIS is to match the coder's placement of minutae, core and deltas, not to 'find the suspect'. Finding the suspect will be a consequence of manually coding the latent in a manner consistent with the automated coding of the reference ten print image.
3) An automated coding of the ten print and latent images alone and launching a search will be roughly 69% accurate. The greatest increase in accuracy comes from manually coding the latent as described in 2) which can result in a +22% accuracy increase. There are other efficiencies such as algorithm upgrades, etc, but they'll roughly get you ~4%.
Oh, I forgot one more point. That is that the rarity of the minutiae coded. Obviously, the rarity of ridge endings terminating within the pattern area of a left slant loop has the potential to produce many more 'close' candidates (I use the term loosely) than coding the minutiae around the core of a nuton loop.
Boyd's comments are accurate. I do want to point out that even if the examiner has an average latent print with 15 points and an average tenprint print that an examiner coded correctly, the system may not produce the suspect or produce it so low on the suspect list it is not evaluated. Thanks for the coment Boyd.
Sure,Boyd also gives us some more tips that be must important.
Really, this blog clearly identify AFlS Accuracy
latent fingerprint kit
latent fingerprint kit
latent fingerprint kit
latent fingerprint kit
latent fingerprint kit
Post a Comment