In looking at the "Executive Summary Strengthening Forensic Science in the United
States: A Path Forward" I felt there was a glaring omission. While the report dealt with the inter-operability of the automated fingerprint identifications it did not address the database issues that should be a major concern with all local and state automated fingerprint identification systems. The poor quality fingerprint arrest card images in the search database, are a significant problem that can lead to missed and bad fingerprint identifications. A missed fingerprint identification allowing a criminal to avoid incarceration is a public safety issue. In the past this has resulted in additional criminal offenses to include homicide.
The reason that poor quality arrest card fingerprints are such a significant problem is because the number of missed and bad fingerprint identifications are significantly greater than the bad latent identifications and often directly impact regular citizens.
While the focus of this report was on the latent fingerprints, the number of useful latent fingerprints at a crime scene pales in comparison to the number of fingerprint cards processed in the course of a year. The FBI recently concluded that even when functioning well the IAFIS fingerprint system is 95% to 98% accurate. I would suggest, a realistic accuracy rate for state and local fingerprint systems would probably be closer to 90%. When this accuracy is combined with the large number of arrest and applicant fingerprint cards, you can see we are dealing with a large number of errors. Poor quality hardcopy fingerprint cards (criminal/non-criminal) increase the potential of error for the journeyman latent fingerprint examiners. A latent print is very often a poor quality image, when compared to a poor quality fingerprint card it increases the possibility of the latent fingerprint examiner making an incorrect conclusion. Since there is no minimum standard for declaring a latent fingerprint identification, examiners often declare an identification with less than 10 points of identification. 10 points, is the number of level II points of identification agreed to by all the examiners in the Mayfield bad identification. Again as noted in an earlier Blog all the latent fingerprint examiners in the Mayfield case, had significant training and experience.
The summary does have some good and important suggestions, which once implemented will improve the reliability of latent fingerprint identifications. The automated fingerprint database acceptance of poor quality fingerprint card images, will continue to produce both missed and bad identification arrest cards.
Bob McAuley
Dir. Operations/Training
Forensic Biometric Identification Solutions LLC.
Showing posts with label fingerprint. Show all posts
Showing posts with label fingerprint. Show all posts
Saturday, February 21, 2009
Monday, February 16, 2009
Madrid Bombing Bad Identification and it's Impact on current identification Procedures
There is no question the FBI latent fingerprint examiners as well as the court appointed latent fingerprint examiner had more training and experience than most fingerprint examiners employed in state and city identification laboratories. In looking at the report generated by the Office of the Inspector General (OIG) the reasons cited for the bad identification are very disturbing and should make every latent fingerprint examiner re-evaluate the criteria they use to make a determination of identification.
The report found " FBI fingerprint examiners relied in significant part on the relationship of "minutiae" or "points" within the prints. These points are places where individual ridges in the fingerprint end or split (ending ridges or bifurcations). These 10 features in Latent Finger Print 17 formed a constellation of points that was generally consistent with the constellation of points in the known fingerprints of both Mayfield and Daoud. The unusual similarity is reflected in the relative location of the points, the orientation of the ridges coming into the points, and the number of intervening ridges between the points. Although the OIG found no systematic study of the rarity of such an event, anecdotal reports suggest that this degree of similarity between prints from two different people is an extremely unusual circumstance."
As every latent fingerprint examiner knows, the minutiae/points is the level II detail. In the 70's as a latent fingerprint examiner, my agency required a minimum number of minutiae or points before a latent fingerprint examiner was allowed to testify in court to an identification. Many other agencies had similar policies. The question that this case brings up is what is the reliability of an identification with 10 or fewer minutiae or points ? I realize a lot has to do with the clarity and uniqueness of the minutiae, but when you're dealing with 10 or fewer points they are usually poor/marginal quality images.
The bias from the examplar tenprint card. " The OIG found that a significant cause of the misidentification was that the Latent Print Unit fingerprint examiners' interpretation of some features in Latent Finger Print 17 was adjusted or influenced by reasoning "backward" from features that were visible in the known fingerprints of Mayfield. This bias is sometimes referred to as "circular reasoning," and is an important pitfall to be avoided. Having found as many as 10 points of unusual similarity, the FBI fingerprint examiners began to "find" additional features in Latent Finger Print 17 that were not really there, but rather were suggested to the latent fingerprint examiners by features in the Mayfield prints. As a result of this process, murky or ambiguous details in Latent Finger Print 17 were erroneously identified as points of similarity with Mayfield's prints."
I'd like to think an experienced latent fingerprint examiner would not allow this bias to influence a decision, but even 4 experienced latent fingerprint examiners were influenced into a bad identification. The most effective way to guard against this would be to have an unbiased 3rd. party latent fingerprint examiner analyze the fingerprint without any knowledge of previous determinations.
Level III detail, "The OIG also found that the FBI latent fingerprint examiners gave significant weight to the purported agreement between extremely tiny details in Latent Finger Print 17 and Mayfield's fingerprints. These details, including shapes interpreted as individual pores, incipient dots between ridges, and ridge edges, are known as "Level 3" details. Because Level 3 details are so small, the appearance of such details in fingerprints is highly variable, even between different fingerprints made by the same finger. As a result, the reliability of Level 3 details is the subject of some controversy within the latent fingerprint community. The OIG found that none of the purported Level 3 features in Latent Finger Print 17 used to identify Mayfield corresponded to features in the known fingerprints of the true donor (Daoud). Thus, unlike the case with larger details, the examiners were not confused by any unusual similarity in Level 3 details on the fingerprints of Mayfield and Daoud. Rather, they apparently misinterpreted distortions in Latent Finger Print 17 as real features corresponding to Level 3 details seen in Mayfield's known fingerprints."
As a latent fingerprint examiner, I rely on Level II detail to make my determination of identification, I do use some Level III detail (ridge shape, etc.) in conjunction with the Level II (ending ridges, bifurcations, etc.) detail, during my comparison and evaluation of the prints. I have not had a situation where my determination of identification relied solely on Level III detail.
One discrepancy rule is followed by all latent examiners, if there is an unexplainable difference in the latent and tenprint fingerprint card, it must be declared a non-identification. "The FBI recognized that the entire upper left portion of Latent Finger Print 17 did not correspond with Mayfield's fingerprint. The examiners explained this difference as being the result of a separate touch, possibly by a different finger or a different person. This explanation required the examiners to accept an extraordinary set of coincidences. The OIG found that the support for this explanation was, at best, contradictory fingerprint examiners are governed by the "one discrepancy rule" in which a single difference in appearance between a latent print fingerprint and a known fingerprint must preclude an identification unless the examiner has a valid explanation for the difference. Latent fingerprint identifications are subject to a standard of 100 percent certainty. Implicit in this standard is the requirement that the examiner have equivalent certainty in the validity of each explanation for each difference in appearance between prints."
"The OIG also found that the FBI examiners failed to give adequate consideration to the incomplete nature of the agreement in points between Latent Finger Print 17 and Mayfield's fingerprint. As previously described, there was a constellation of as many as 10 points in LFP 17 that bore an unusual similarity to points in the Mayfield fingerprint. However, the limited clarity of Latent Finger Print 17 prevented, the latent fingerprint examiners from making an accurate determination of the type of many of these points (that is, whether they were ending ridges or bifurcations). The OIG found that many of the points utilized by the FBI to support the identification suffered from this shortcoming (ambiguity as to feature type), and that accordingly the "quality" of the agreement was inadequate to support the conclusion of identification." Apparently the latent fingerprint examiners allowed themselves to be influenced by each other and the tenprint fingerprint card.
What is clear from the Madrid bombing error is that even highly trained and experienced latent fingerprint examiners can make identification errors. Most latent fingerprint examiners testifying in court today, don't have near the training or experience of the 4 individuals above and often the identification is based on the opinion of only 1 latent fingerprint examiner. The automated fingerprint identification systems today are designed to find candidate fingerprints having the most minutiae arrangements similar to the encoded minutiae from the latent fingerprint. These candidates should include the correct match of the print (if it is in the database), but will also include the closest possible non-matches. These close non-match suspects require more fingerprint comparison experience than many of today's latent examiners have been able to acquire.
The report found " FBI fingerprint examiners relied in significant part on the relationship of "minutiae" or "points" within the prints. These points are places where individual ridges in the fingerprint end or split (ending ridges or bifurcations). These 10 features in Latent Finger Print 17 formed a constellation of points that was generally consistent with the constellation of points in the known fingerprints of both Mayfield and Daoud. The unusual similarity is reflected in the relative location of the points, the orientation of the ridges coming into the points, and the number of intervening ridges between the points. Although the OIG found no systematic study of the rarity of such an event, anecdotal reports suggest that this degree of similarity between prints from two different people is an extremely unusual circumstance."
As every latent fingerprint examiner knows, the minutiae/points is the level II detail. In the 70's as a latent fingerprint examiner, my agency required a minimum number of minutiae or points before a latent fingerprint examiner was allowed to testify in court to an identification. Many other agencies had similar policies. The question that this case brings up is what is the reliability of an identification with 10 or fewer minutiae or points ? I realize a lot has to do with the clarity and uniqueness of the minutiae, but when you're dealing with 10 or fewer points they are usually poor/marginal quality images.
The bias from the examplar tenprint card. " The OIG found that a significant cause of the misidentification was that the Latent Print Unit fingerprint examiners' interpretation of some features in Latent Finger Print 17 was adjusted or influenced by reasoning "backward" from features that were visible in the known fingerprints of Mayfield. This bias is sometimes referred to as "circular reasoning," and is an important pitfall to be avoided. Having found as many as 10 points of unusual similarity, the FBI fingerprint examiners began to "find" additional features in Latent Finger Print 17 that were not really there, but rather were suggested to the latent fingerprint examiners by features in the Mayfield prints. As a result of this process, murky or ambiguous details in Latent Finger Print 17 were erroneously identified as points of similarity with Mayfield's prints."
I'd like to think an experienced latent fingerprint examiner would not allow this bias to influence a decision, but even 4 experienced latent fingerprint examiners were influenced into a bad identification. The most effective way to guard against this would be to have an unbiased 3rd. party latent fingerprint examiner analyze the fingerprint without any knowledge of previous determinations.
Level III detail, "The OIG also found that the FBI latent fingerprint examiners gave significant weight to the purported agreement between extremely tiny details in Latent Finger Print 17 and Mayfield's fingerprints. These details, including shapes interpreted as individual pores, incipient dots between ridges, and ridge edges, are known as "Level 3" details. Because Level 3 details are so small, the appearance of such details in fingerprints is highly variable, even between different fingerprints made by the same finger. As a result, the reliability of Level 3 details is the subject of some controversy within the latent fingerprint community. The OIG found that none of the purported Level 3 features in Latent Finger Print 17 used to identify Mayfield corresponded to features in the known fingerprints of the true donor (Daoud). Thus, unlike the case with larger details, the examiners were not confused by any unusual similarity in Level 3 details on the fingerprints of Mayfield and Daoud. Rather, they apparently misinterpreted distortions in Latent Finger Print 17 as real features corresponding to Level 3 details seen in Mayfield's known fingerprints."
As a latent fingerprint examiner, I rely on Level II detail to make my determination of identification, I do use some Level III detail (ridge shape, etc.) in conjunction with the Level II (ending ridges, bifurcations, etc.) detail, during my comparison and evaluation of the prints. I have not had a situation where my determination of identification relied solely on Level III detail.
One discrepancy rule is followed by all latent examiners, if there is an unexplainable difference in the latent and tenprint fingerprint card, it must be declared a non-identification. "The FBI recognized that the entire upper left portion of Latent Finger Print 17 did not correspond with Mayfield's fingerprint. The examiners explained this difference as being the result of a separate touch, possibly by a different finger or a different person. This explanation required the examiners to accept an extraordinary set of coincidences. The OIG found that the support for this explanation was, at best, contradictory fingerprint examiners are governed by the "one discrepancy rule" in which a single difference in appearance between a latent print fingerprint and a known fingerprint must preclude an identification unless the examiner has a valid explanation for the difference. Latent fingerprint identifications are subject to a standard of 100 percent certainty. Implicit in this standard is the requirement that the examiner have equivalent certainty in the validity of each explanation for each difference in appearance between prints."
"The OIG also found that the FBI examiners failed to give adequate consideration to the incomplete nature of the agreement in points between Latent Finger Print 17 and Mayfield's fingerprint. As previously described, there was a constellation of as many as 10 points in LFP 17 that bore an unusual similarity to points in the Mayfield fingerprint. However, the limited clarity of Latent Finger Print 17 prevented, the latent fingerprint examiners from making an accurate determination of the type of many of these points (that is, whether they were ending ridges or bifurcations). The OIG found that many of the points utilized by the FBI to support the identification suffered from this shortcoming (ambiguity as to feature type), and that accordingly the "quality" of the agreement was inadequate to support the conclusion of identification." Apparently the latent fingerprint examiners allowed themselves to be influenced by each other and the tenprint fingerprint card.
What is clear from the Madrid bombing error is that even highly trained and experienced latent fingerprint examiners can make identification errors. Most latent fingerprint examiners testifying in court today, don't have near the training or experience of the 4 individuals above and often the identification is based on the opinion of only 1 latent fingerprint examiner. The automated fingerprint identification systems today are designed to find candidate fingerprints having the most minutiae arrangements similar to the encoded minutiae from the latent fingerprint. These candidates should include the correct match of the print (if it is in the database), but will also include the closest possible non-matches. These close non-match suspects require more fingerprint comparison experience than many of today's latent examiners have been able to acquire.
Bob McAuley
Dir. Operations/Training
Forensic Biometric Identification Solutions LLC.
Tuesday, February 10, 2009
"It takes 10 years of extensive training to excel in anything" Herbert Simon - Nobel Laureate
Scientific research has concluded that it takes eight-to-twelve years of training for a talented player/athlete to reach elite levels. This is called the ten-year or 10,000 hour rule, which translates to slightly more than three hours of practice daily for ten years (Ericsson, et al., 1993; Ericsson and Charness, 1994, Bloom, 1985; Salmela et al., 1998).
Does this apply to latent fingerprint examiners? and how does this impact latent fingerprint accuracy? I think everyone will agree that today latent fingerprint examiners have the opportunity to identify many more latent fingerprints than in the past, largely due to the automated fingerprint identification systems. I would estimate that 20 to 30% of the cases identified in my Lab. were cold hits ( fingerprint suspects produced by the automated fingerprint identification system). In addition to the increase in latent fingerprint identifications, there has been a decrease in image quality of the fingerprint cards housed in state and federal databases (tenprint cards). Many fingerprint cards are still poorly acquired and the fingerprint card printers on the automated systems often don't produce a high quality image for fingerprint comparison. What does that mean for the latent fingerprint examiner? it means that today there are many more difficult latent fingerprint comparisons for the latent examiner. Are there more missed and bad fingerprint identifications today, unfortunately, yes. In addition to the increase in the number of latent fingerprints developed and searched , the automated fingerprint identification system computers find unusually close non-matching fingerprint suspects. To make a determination of identification today it will require a latent fingerprint examiner that has reached " elite levels", the individual with 10,000 hrs of fingerprint comparison work and training. Unfortunately many of the local agencies don't have the resources to develop the latent fingerprint examiner to the elite levels, often they develop a fingerprint examiner that is good at many functions (a jack of all trades). There are many fingerprint identifications that can be done by journeymen fingerprint examiners. When the latent fingerprint lift and fingerprint card are both good, the fingerprint examiner with limited experience is capable making a good decision. The problem for the journeyman fingerprint examiner is when the print lacks clarity and they are required to determine a level of tolerance for discrepancies. The answer of course is more fingerprint experience and training before the latent examiner is allowed to testify in court. Is the problem of more fingerprint comparison work before allowing the fingerprint examiner to testify in court being addressed? Unfortunately it is not and we can expect more Boston PD and L.A. PD Lab. situations from other agencies in the future. Some fingerprint programs that do appear to work are the FBI and the larger states which do allow the latent examiner to reach the elite levels of experience before they are allowed to testify in court. The FBI requires that latent fingerprint examiners make in the neighborhood of 250,000 comparisons before they are ready to testify in court. The larger state agencies, New York as an example, develop fingerprint examiners through the tenprint system as part of a career ladder. It usually takes a fingerprint examiner close to 10 yrs to work up through the system to get a chance to be a latent fingerprint examiner. Both of these systems produce excellent latent fingerprint examiners. Most PD's and smaller states, because of budgetary issues, aren't able to develop elite level latent fingerprint examiners and will become the LA or Boston storyies in the future.
Does this apply to latent fingerprint examiners? and how does this impact latent fingerprint accuracy? I think everyone will agree that today latent fingerprint examiners have the opportunity to identify many more latent fingerprints than in the past, largely due to the automated fingerprint identification systems. I would estimate that 20 to 30% of the cases identified in my Lab. were cold hits ( fingerprint suspects produced by the automated fingerprint identification system). In addition to the increase in latent fingerprint identifications, there has been a decrease in image quality of the fingerprint cards housed in state and federal databases (tenprint cards). Many fingerprint cards are still poorly acquired and the fingerprint card printers on the automated systems often don't produce a high quality image for fingerprint comparison. What does that mean for the latent fingerprint examiner? it means that today there are many more difficult latent fingerprint comparisons for the latent examiner. Are there more missed and bad fingerprint identifications today, unfortunately, yes. In addition to the increase in the number of latent fingerprints developed and searched , the automated fingerprint identification system computers find unusually close non-matching fingerprint suspects. To make a determination of identification today it will require a latent fingerprint examiner that has reached " elite levels", the individual with 10,000 hrs of fingerprint comparison work and training. Unfortunately many of the local agencies don't have the resources to develop the latent fingerprint examiner to the elite levels, often they develop a fingerprint examiner that is good at many functions (a jack of all trades). There are many fingerprint identifications that can be done by journeymen fingerprint examiners. When the latent fingerprint lift and fingerprint card are both good, the fingerprint examiner with limited experience is capable making a good decision. The problem for the journeyman fingerprint examiner is when the print lacks clarity and they are required to determine a level of tolerance for discrepancies. The answer of course is more fingerprint experience and training before the latent examiner is allowed to testify in court. Is the problem of more fingerprint comparison work before allowing the fingerprint examiner to testify in court being addressed? Unfortunately it is not and we can expect more Boston PD and L.A. PD Lab. situations from other agencies in the future. Some fingerprint programs that do appear to work are the FBI and the larger states which do allow the latent examiner to reach the elite levels of experience before they are allowed to testify in court. The FBI requires that latent fingerprint examiners make in the neighborhood of 250,000 comparisons before they are ready to testify in court. The larger state agencies, New York as an example, develop fingerprint examiners through the tenprint system as part of a career ladder. It usually takes a fingerprint examiner close to 10 yrs to work up through the system to get a chance to be a latent fingerprint examiner. Both of these systems produce excellent latent fingerprint examiners. Most PD's and smaller states, because of budgetary issues, aren't able to develop elite level latent fingerprint examiners and will become the LA or Boston storyies in the future.
Friday, February 6, 2009
Brady vs. Maryland and Giglio vs. United States
Brady vs. Maryland and Giglio vs. United States make it clear that prosecutors must disclose to the defense information which bears on the reliability or veracity of witnesses. U.S. Supreme Court decisions have enforced the "Brady Rule" to include evidence maintained in personnel files. That means that when a latent, tenprint fingerprint examiner or crime scene specialist has intentionally communicated either verbally or in writing a deceptive documented statement or message, the prosecutor must inform the defense. Intentional deceptive action in a formal setting, such as testifying in court, during an internal affairs investigation or false evidence that tends to implicate another in a criminal act, should result in termination or permanent removal from any position where the individual would be called upon to be a witness in any court action. The situation becomes less clear when the individual's actions cannot be determined to be intentional. An internal investigation might determine there was no evidence of criminal intent. An example might be an individual makes a wrong determination on a fingerprint identity, or crime scene tech. associates the wrong individual(s) with the scene based on the evidence collected. It is clear that the individual's history would be discoverable and impact the credibility of that individual testimony. What is not clear is the impact on a unit if the supervisor of the unit has the credibility issue, example the Mayfield case. If the testimony will come from another individual in the unit, does the fact that the supervisor may be perceived as having a credibility issue have any bearing on the case ?
Sunday, February 1, 2009
Migration from 500ppi to 1,000ppi Fingerprint Search Database
While the decision to go to 1,000 ppi on fingerprint card capture was made some time ago it would appear that the bad economy has slowed, if not completely halted the migration. Most states have not migrated to 1,000 ppi. fingerprint card capture, so there are only a few states that find themselves in a situation where they are running automated fingerprint identification systems with search files consisting of both 500 ppi and 1,000 ppi fingerprint card images. The FBI does plan to move to 1,000 ppi fingerprint cards at some point in the future, but I would expect, that they are unable to make the move until all the large states have migrated to the 1,000 ppi fingerprint card capture systems. The issues confronting states from migrating will be cost issue's:
The FBI currently will accept the 1,000 ppi fingerprint images but they are storing the high resolution fingerprint images; they downsize the fingerprint image for searching and updating to the FBI fingerprint search database. So in effect the states submitting 1,000 ppi. fingerprint images are actually getting a 500 ppi. fingerprint search. This I would suspect was originally planned as a temporary solution by the FBI until all states moved to the higher resolution fingerprint systems.
So lets discuss some of the concerns with the migration from 500 to the 1,000 ppi fingerprint system. While getting a higher resolution fingerprint image should give agencies a more accurate fingerprint search is a logical assumption, the question is with the interim solution migrating to the higher resolution system will this assumption be true ? Running a system with fingerprint search database of 500ppi. and 1,000ppi. fingerprint images will create challenges for the fingerprint coders and matchers. The images are not the same size and they each are compressed using different compression algorithms with different ratios of compression. Will the interim fingerprint search system be able to match the current fingerprint system accuracy? The interim system would capture new fingerprint images at 1K ppi. and if an identification is made, replace the lower resolution fingerprint image with the new higher resolution fingerprint image. It is possible some states could manually convert all the hardcopy fingerprint cards in their database, reducing the number of lower resolution fingerprint cards left in the database as well as reducing the length of time they run the dual system. Remember so many fingerprint cards have been captured electronically over the last few years and these fingerprint images captured at 500 ppi cannot be interpolated up to 1,000 ppi. A big question is will the minutiae extraction algorithms and matcher algorithms be affected by having a mixed resolution database ? Each vendor,each agency, each version of afis software may be impacted a little or a lot. Remember the fingerprint minutiae extraction programs were developed for 500 ppi images and not for a dual resolution fingerprint systems. This is a question that has not been answered. I have not heard of any fingerprint system accuracy tests involving fingerprint systems with dual resolution fingerprint databases. I would also point out that most vendors test in a controlled situation with good quality fingerprint images to get the high fingerprint accuracy rates. In the real world the fingerprint accuracy rates are significantly lower because of poorly taken tenprints and partial latent images. In my experience the hit rate for latent fingerprint searches is seldom above 15%, this percentage is based on the number of unknown latent fingerprints entered into the system that are identified. Every examiner knows that getting a latent fingerprint hit is more dependent on the quality of the tenprint fingerprint image in the database than anything else. So whether the fingerprint card is captured at 500 ppi. or at 1K ppi the fingerprint accuracy, will be dependent on the quality of the fingerprint images captured at the livescan unit and how the matchers and coders handle dual fingerprint resolutions.
- Bandwidth for the higher resolution fingerprint card images.
- Conversion costs.
- Fingerprint livescan units are twice as expensive for the higher resolution units.
- System maintenance costs are a % of system cost so there is significant increase.
The FBI currently will accept the 1,000 ppi fingerprint images but they are storing the high resolution fingerprint images; they downsize the fingerprint image for searching and updating to the FBI fingerprint search database. So in effect the states submitting 1,000 ppi. fingerprint images are actually getting a 500 ppi. fingerprint search. This I would suspect was originally planned as a temporary solution by the FBI until all states moved to the higher resolution fingerprint systems.
So lets discuss some of the concerns with the migration from 500 to the 1,000 ppi fingerprint system. While getting a higher resolution fingerprint image should give agencies a more accurate fingerprint search is a logical assumption, the question is with the interim solution migrating to the higher resolution system will this assumption be true ? Running a system with fingerprint search database of 500ppi. and 1,000ppi. fingerprint images will create challenges for the fingerprint coders and matchers. The images are not the same size and they each are compressed using different compression algorithms with different ratios of compression. Will the interim fingerprint search system be able to match the current fingerprint system accuracy? The interim system would capture new fingerprint images at 1K ppi. and if an identification is made, replace the lower resolution fingerprint image with the new higher resolution fingerprint image. It is possible some states could manually convert all the hardcopy fingerprint cards in their database, reducing the number of lower resolution fingerprint cards left in the database as well as reducing the length of time they run the dual system. Remember so many fingerprint cards have been captured electronically over the last few years and these fingerprint images captured at 500 ppi cannot be interpolated up to 1,000 ppi. A big question is will the minutiae extraction algorithms and matcher algorithms be affected by having a mixed resolution database ? Each vendor,each agency, each version of afis software may be impacted a little or a lot. Remember the fingerprint minutiae extraction programs were developed for 500 ppi images and not for a dual resolution fingerprint systems. This is a question that has not been answered. I have not heard of any fingerprint system accuracy tests involving fingerprint systems with dual resolution fingerprint databases. I would also point out that most vendors test in a controlled situation with good quality fingerprint images to get the high fingerprint accuracy rates. In the real world the fingerprint accuracy rates are significantly lower because of poorly taken tenprints and partial latent images. In my experience the hit rate for latent fingerprint searches is seldom above 15%, this percentage is based on the number of unknown latent fingerprints entered into the system that are identified. Every examiner knows that getting a latent fingerprint hit is more dependent on the quality of the tenprint fingerprint image in the database than anything else. So whether the fingerprint card is captured at 500 ppi. or at 1K ppi the fingerprint accuracy, will be dependent on the quality of the fingerprint images captured at the livescan unit and how the matchers and coders handle dual fingerprint resolutions.
Wednesday, January 28, 2009
Fingerprint Search Database, not all created equal
The make up of today’s fingerprint search databases have evolved from databases primarily made up of criminal fingerprint cards, to systems that now include significant numbers of non-criminal applicant and permit fingerprint cards. These criminal background checks have been a tremendous benefit to public safety and in all likelihood prevented numerous crimes, some of which would have led to serious injury or death. As the number of non-criminal records increase it becomes even more critical that the fingerprint search databases and procedures adhere to the highest standards to decrease the possibility of error. The Brandon Mayfield case in Oregon is probably the most recognizable case where an individual was wrongly identified using fingerprint. There are quite a number of other cases, the individual stopped in CA. for a traffic infraction but based on a fingerprint criminal history, was wrongly incarcerated. In the lawsuits that followed, it was found the error had occurred 4 years earlier and the state was aware of 97 other similar identification errors. The cause of the error was attributed to hardware problem. As is apparent from the last example, the error may not become apparent until years later and can be difficult to get corrected.
Are the images in all the state search databases the same, unfortunately, they are not, the fingerprints in each state search database are dictated by state statute. Every state requires felony arrests be captured and submitted and stored in the state repository, not all states require all misdemeanors to be fingerprinted and stored at the state level. The non-criminal applicant and permit cards are also dictated by state statute, with some non-criminal cards retained by the state and some returned to the submitting agency. The FBI and NY state have specific guidelines on what they can accept and store in their respective search databases. When a fingerprint card is processed that cannot be retained the software insures that after processing the electronic images and alpha-numeric data are not retained in the search database. Not all states adhere to these standards and in some situations non-criminal cards after processing are to the submitting agency, the electronic images and alpha-numeric information are stored in the search database and become part of the fingerprint searche database. I’m sure if the legislature required the card be returned, their intent would be the electronic copies of fingerprint images and alpha-numeric information be destroyed.The questions that this brings up are:
1) A latent print is identified to a fingerprint that there is no state statute requiring the state to store the print in the search database, will the court accept the print or toss the evidence?
2) An error occurs and a criminal history is associated with the non-criminal history, that should have been returned, is the state liable ?
Lets look at the fingerprint images and how they are stored in the search database. The images stored could be:
A) the original fingerprint event creating the record.
B) a composite image, the search iamges are created from the best images from multiple cards.
C) All the fingerprint cards that come in are stored are stored in the search database.
Just to add a little more diversity to the database search fingerprint images they can be stored at 500ppi, 1000ppi or a combination at both resolutions.Lets look at how the images are stored:
A) The FBI database uses the originating fingerprint card, the 1st. event. If subsequent cards have better quality images they update the new minutiae (points of identification used in the search), a good way to process.
B) The composite image database is used by many states and if done properly with the policies and procedures can work well, but it is more prone to error than the other methods. If for example you replace a image in the search database from a non-retainable card.
C) The systems that have all the cards available in the search database, probably the best way to store the search images but there would be cost considerations.No matter what system is used the database search images should periodically be reconciled, as part of the system regular maintenance, it is just as critical as backing up the system.
I also will talk about the effect of multiple resolution systems and especially the ones with a composite.The resolution questions will probably be in the next blog.
Are the images in all the state search databases the same, unfortunately, they are not, the fingerprints in each state search database are dictated by state statute. Every state requires felony arrests be captured and submitted and stored in the state repository, not all states require all misdemeanors to be fingerprinted and stored at the state level. The non-criminal applicant and permit cards are also dictated by state statute, with some non-criminal cards retained by the state and some returned to the submitting agency. The FBI and NY state have specific guidelines on what they can accept and store in their respective search databases. When a fingerprint card is processed that cannot be retained the software insures that after processing the electronic images and alpha-numeric data are not retained in the search database. Not all states adhere to these standards and in some situations non-criminal cards after processing are to the submitting agency, the electronic images and alpha-numeric information are stored in the search database and become part of the fingerprint searche database. I’m sure if the legislature required the card be returned, their intent would be the electronic copies of fingerprint images and alpha-numeric information be destroyed.The questions that this brings up are:
1) A latent print is identified to a fingerprint that there is no state statute requiring the state to store the print in the search database, will the court accept the print or toss the evidence?
2) An error occurs and a criminal history is associated with the non-criminal history, that should have been returned, is the state liable ?
Lets look at the fingerprint images and how they are stored in the search database. The images stored could be:
A) the original fingerprint event creating the record.
B) a composite image, the search iamges are created from the best images from multiple cards.
C) All the fingerprint cards that come in are stored are stored in the search database.
Just to add a little more diversity to the database search fingerprint images they can be stored at 500ppi, 1000ppi or a combination at both resolutions.Lets look at how the images are stored:
A) The FBI database uses the originating fingerprint card, the 1st. event. If subsequent cards have better quality images they update the new minutiae (points of identification used in the search), a good way to process.
B) The composite image database is used by many states and if done properly with the policies and procedures can work well, but it is more prone to error than the other methods. If for example you replace a image in the search database from a non-retainable card.
C) The systems that have all the cards available in the search database, probably the best way to store the search images but there would be cost considerations.No matter what system is used the database search images should periodically be reconciled, as part of the system regular maintenance, it is just as critical as backing up the system.
I also will talk about the effect of multiple resolution systems and especially the ones with a composite.The resolution questions will probably be in the next blog.
Sunday, January 25, 2009
No matter how good a system purchased the following factors will determine accuracy
In my earlier blog, I discussed accuracy issues inherent in all automated fingerprint identification systems. Now I’d like to discuss what many states fail to realize, system accuracy it is not based solely on the hardware purchased. The factors that most influence system accuracy and performance are usually determined by the choices made by the agency representative(s) during the development of the RFP. There is no doubt that the FBI system is the gold standard in automated fingerprint identification systems, based on the high standards of fingerprint quality they can require for submitted fingerprint cards. Lets look at some of the other factors which significantly impact system accuracy:
The training and experience of the individuals processing the arrest cards. The FBI and NYS have excellent training programs which include in-service, proficiency and remediation training, after the ten print examiner has completed a basic pattern recognition/verification course. Not all states have formalized training programs with many states only requiring the basic course. Agencies should be pro-active in requiring training( in-service, proficiency and remediation programs) to insure the work being produced and entered into the database is accurate.
Staffing and work flow is another area that will impact the accuracy of a system. State production standards and work flows are designed to get the maximum number of fingerprint cards through the system as fast as possible. This means that many systems are designed to default to a hit/miss determination with just a click of the mouse, not requiring the examiner to pause before making the critical decision. An examiner pushing to surpass production targets is prone to error. Contrast this with a latent examiner who can spend a few minutes or a few hours making a determination on a possible identification.
Single vs. double verification of an identification. The FBI and NYS systems initially used a single verification when processing fingerprint cards in the automated fingerprint identifications systems. Both systems moved to a double verification after the error rate using the single verification procedure led to dramatic increases in bad identifications. Many states however rely on a single verification in ten-print processing, these systems are prone to error. I would also point out that as states continue to increase the number of non-criminal cards processed, $ liability exposure for the state increases.
Database maintenance, the FBI makes available to the state records that may be discrepant. While these discrepant record lists are useful, not all states follow-up and reconcile these records. Since the FBI does not have all the state cards (poor quality and many app’s) the FBI list is only a partial reconciliation list. States should have a database reconciliation plan to insure the integrity of the database records. I know of one system where there were 1,200 records with only fingerprint images, no alpha-numeric data (no name,dob,date fingerprinted, reason fingerprinted) but the fingerprints were part of the search database. NY state by contrast has a program that runs regularly scheduled computerized checks of the database to locate and correct discrepant records, in addition to monthly reconciliation with the FBI list of discrepant records.
What I am amazed at is there are no minimum training standards from the FBI for state ten-print examiners, these examiners through the state system are actually entering fingerprints into the national database. All of the factors described above could be addressed at very little, if any additional, cost to agencies. The training, proficiency and remediation costs for ten-print examiners could be handled by any agencies current training budget, it would only require a knowledgeable person setting it up the training/remediation programs. The double verification at agencies with a small staff could be accomplished using lights out technology, and slight adjustment to the current work flow. Agencies pushing production, need to realize how consequential ten print errors can be, the devastating and dramatic effect on public safety as well as the liability exposure to the agency. We’ll discuss latent searches in more detail in the future, but remember the ten print database integrity is critical to latent searching and accuracy (GIGO).
Note, what I have found is an agency that utilizes people with a limited knowledge of how the automated fingerprint identification system works, in the end, spend more money on a more unreliable system.
The training and experience of the individuals processing the arrest cards. The FBI and NYS have excellent training programs which include in-service, proficiency and remediation training, after the ten print examiner has completed a basic pattern recognition/verification course. Not all states have formalized training programs with many states only requiring the basic course. Agencies should be pro-active in requiring training( in-service, proficiency and remediation programs) to insure the work being produced and entered into the database is accurate.
Staffing and work flow is another area that will impact the accuracy of a system. State production standards and work flows are designed to get the maximum number of fingerprint cards through the system as fast as possible. This means that many systems are designed to default to a hit/miss determination with just a click of the mouse, not requiring the examiner to pause before making the critical decision. An examiner pushing to surpass production targets is prone to error. Contrast this with a latent examiner who can spend a few minutes or a few hours making a determination on a possible identification.
Single vs. double verification of an identification. The FBI and NYS systems initially used a single verification when processing fingerprint cards in the automated fingerprint identifications systems. Both systems moved to a double verification after the error rate using the single verification procedure led to dramatic increases in bad identifications. Many states however rely on a single verification in ten-print processing, these systems are prone to error. I would also point out that as states continue to increase the number of non-criminal cards processed, $ liability exposure for the state increases.
Database maintenance, the FBI makes available to the state records that may be discrepant. While these discrepant record lists are useful, not all states follow-up and reconcile these records. Since the FBI does not have all the state cards (poor quality and many app’s) the FBI list is only a partial reconciliation list. States should have a database reconciliation plan to insure the integrity of the database records. I know of one system where there were 1,200 records with only fingerprint images, no alpha-numeric data (no name,dob,date fingerprinted, reason fingerprinted) but the fingerprints were part of the search database. NY state by contrast has a program that runs regularly scheduled computerized checks of the database to locate and correct discrepant records, in addition to monthly reconciliation with the FBI list of discrepant records.
What I am amazed at is there are no minimum training standards from the FBI for state ten-print examiners, these examiners through the state system are actually entering fingerprints into the national database. All of the factors described above could be addressed at very little, if any additional, cost to agencies. The training, proficiency and remediation costs for ten-print examiners could be handled by any agencies current training budget, it would only require a knowledgeable person setting it up the training/remediation programs. The double verification at agencies with a small staff could be accomplished using lights out technology, and slight adjustment to the current work flow. Agencies pushing production, need to realize how consequential ten print errors can be, the devastating and dramatic effect on public safety as well as the liability exposure to the agency. We’ll discuss latent searches in more detail in the future, but remember the ten print database integrity is critical to latent searching and accuracy (GIGO).
Note, what I have found is an agency that utilizes people with a limited knowledge of how the automated fingerprint identification system works, in the end, spend more money on a more unreliable system.
Monday, January 19, 2009
FBI checking database to make sure it is matching fugitives' fingerprints
The above headline was from an article on the case of Jeremy Jones whose initial Georgia arrest when processed failed to match his fingerprints to a prior fingerprint record in Oklahoma. When a match was not made, a new file in the FBI database was created for "Chapman". Jones was wanted in Oklahoma for jumping bail in 2000, where he was charged with two counts of rape and two counts of sodomy. After the initial missed identification Jones was responsible for 3 homicides. After his arrest he unexpectedly made a startling confession, he also confessed to 13 other murders across six states. The FBI review concluded that even when functioning well, the fingerprint system is 95% to 98% accurate, as it process's 50,000 prints a day. This would mean that the FBI could possibly be missing 1,000 records a day. FBI computer technicians could adjust the computer to produce more potential fingerprint matches in more cases."But that will cost law enforcement time and give you more false positives,".The thing I want to point out is that fingerprint arrest card processing is critical to a good state system and even the best systems produce errors, the consequences from a fingerprint error as you see can be significant. While fingerprint errors resulting in homicides are rare, I know of two (2) other instances where a tenprint fingerprint miss has resulted in additional homicide victims. Usually a missed fingerprint identification results in incorrect criminal history information being associated with an individual, which often results in liability costs for the state. The FBI fingerprint system does have a significant advantage over most state agency fingerprint systems when it comes to accuracy, it can reject poor quality or improperly rolled fingerprint arrest cards. Since the states are often required to accept and process poor quality(reprint not available) fingerprint arrest cards, is highly unlikely that a state system could come close to the FBI IAFIS on fingerprint system accuracy. The result is a state fingerprint database that has questionable fingerprint images in the search database, negatively impacting fingerprint accuracy. A very significant problem with these poor quality fingerprint arrest records, the FBI will not accept and process these fingerprint records, so they will not appear on the FBI criminal history, no NCIC record of that fingerprint event.
Many states rely on a train the trainer concept to get booking officers trained properly in the hope of improving fingerprint image quality, unfortunately because of turnover this concept has not been successful. I would suggest that the proper taking of fingerprints training be a part of all states jail standards and that booking officers be required to attend formal training on the proper techniques required to capture a good set of fingerprint images. In addition to the initial training the booking officers should receive periodic formal in-service refresher training on capturing good quality fingerprints as part of their job requirements.The training should cover capturing fingerprints using both ink & roll and livescan technology, even if it requires hiring a private vendor. Maybe just to start another train of thought, an old idea that has been brought up numerous times by just about all latent fingerprint examiners. Why not go back to the full pattern designation in afis ? central pocket loop whorl etc. It is a way to increase fingerprint accuracy and the Jeremy Jones error probably would not have occurred.
Many states rely on a train the trainer concept to get booking officers trained properly in the hope of improving fingerprint image quality, unfortunately because of turnover this concept has not been successful. I would suggest that the proper taking of fingerprints training be a part of all states jail standards and that booking officers be required to attend formal training on the proper techniques required to capture a good set of fingerprint images. In addition to the initial training the booking officers should receive periodic formal in-service refresher training on capturing good quality fingerprints as part of their job requirements.The training should cover capturing fingerprints using both ink & roll and livescan technology, even if it requires hiring a private vendor. Maybe just to start another train of thought, an old idea that has been brought up numerous times by just about all latent fingerprint examiners. Why not go back to the full pattern designation in afis ? central pocket loop whorl etc. It is a way to increase fingerprint accuracy and the Jeremy Jones error probably would not have occurred.
Subscribe to:
Posts (Atom)
Posts
