 |
November 2004, Page 24
Does The Use Of Digital Techniques By Law Enforcement Authorities Create A Risk Of Miscarriages Of Justice?
By Michael Cherry; Edward J. Imwinkelried; Larry Meyer
Approximately 30 million criminal fingerprint submissions were entered into the Automated Fingerprint Identification System (AFIS) and its relatives between October 2002 and May 2003. These are fast, convenient identification systems. Speed and convenience are certainly desirable characteristics, but the most essential characteristic of any forensic technique used in the justice and homeland security systems must be its accuracy. Are these systems accurate? Do they enable fingerprint analysts to correctly single out the person who left the crime scene impression? Will they enable us to identify terrorists?
Some Individual Case Studies
The horror stories of Brandon Mayfield, Roger Benson, and Miguel Espinoza supply a partial answer to these vital questions. The published accounts of these stories are both revealing and troublesome.
Brandon Mayfield was released Thursday, May 27, 2004, after Spanish officials conceded that fingerprints on a bag left near the Madrid bombing site were those of an Algerian, not Mayfield’s. However, earlier Mayfield had been mistakenly detained as the person who left the fingerprints. After the March terrorist attack on commuter trains in Madrid, partial latent fingerprints were discovered and lifted from plastic bags that had contained detonator caps. Digital images of the latents were prepared and submitted by Spanish authorities to our FBI for analysis. (See The Champion, September/October 2004.)
According to the official statement issued by the FBI on May 24, the submitted images were searched through the Integrated Automated Fingerprint Identification System (IAFIS). An IAFIS search compares an unknown print to a database of millions of known prints. The result of the search is a short list of possible matches. A trained fingerprint examiner then takes the short list and performs an examination to determine whether the unknown print matches a known print in the database. In this case, the FBI examiners initially concluded that the latents were Mayfield’s.
“Using standard protocols and methodologies, FBI fingerprint examiners determined that the latent fingerprint was of value for identification purposes. This print was subsequently linked to Brandon Mayfield. That association was independently analyzed, and the results were confirmed by an outside experienced fingerprint expert,” the FBI said. Yet, “upon subsequent review it was determined that the FBI identification was based on an image of substandard quality, which was particularly problematic because of the remarkable number of points of similarity between Mr. Mayfield’s prints and the print details in the images submitted to the FBI.”
It thus appears that rather than directly analyzing the original image of the latent fingerprints, at first the FBI examiners compared IAFIS exemplars to digital images provided by the Spanish government. The use of the digital images probably contributed to the mistake. Since digital images have lower resolution, they contain less detail. If the digital images had included the missing details, the more complete images might have made it evident much sooner to the FBI examiners that the latent prints were not Mayfield’s.
Similarities at low detail (resolution) are not necessarily similarities at higher detail.
Roger Benson’s story compounds the concerns raised by the Mayfield case. Benson was confined to the maximum security “D Block” for 43 days in the Yreka County Jail in California. Newspaper and Web accounts1 indicate that eight months after Benson was assigned an electronic fingerprint number to represent his IAFIS ten finger exemplar, personal information, and arrest history, the very same number was assigned to William Lee Kellogg, a convicted felon.
According to a published account, “Benson’s clean record became drastically corrupted by Kellogg’s criminal record, and the problem was only worsened when a couple of months after his arrest Kellogg was convicted of three felony counts of Criminal Mistreatment in the First Degree. The record mingled accurate physical descriptions of Roger Benson with William Lee Kellogg’s criminal record. As a result, when Benson encountered a police officer during a routine investigation and honestly disclosed that he had a .22 pistol in his truck, he was getting himself into trouble. When the police officer looked at Benson’s record, he was actually reading one of the Livescan Machine’s multiple, destructive forgeries; this forgery said that Benson was a convicted felon, and therefore was not allowed to carry a weapon.”2
Benson’s attorney, Charles Carreon, of Ashland, Oregon knows of more than 90 similar West Coast cases, including that of Miguel Espinoza. Miguel Espinoza was a successful restauranteur in Medford, a town in Lake County, Oregon. Espinoza’s electronic fingerprint number was also assigned to a killer. Consequently, Espinoza was misidentified as a criminal when the city later ran a background check on all liquor license holders. The Oregon State Police computer said that Espinoza had murdered someone. Based on the erroneous background check, the town council revoked his liquor license and destroyed his business.
Some believe that these problems are ancient history and unlikely to recur. Lieutenant Cliff Daimler of the Oregon State Police identification unit stated that the problems have been rectified. “We corrected all 97 [Lane County mix-ups] as soon as we were made aware of them,” he told Williamete Week. “We’re certainly not aware of any issues in other counties.” But OSP spokesman Lieutenant Glenn Chastain conceded that “to verify every single entry that comes in would be impossible.” The Oregon State Police indexing process assigned the same electronic fingerprint number to multiple persons. It stands to reason that if other states use similar indexing systems, the same mistake could have occurred elsewhere. In short, we may not have heard the last horror story.
A Broader Perspective
As Mayfield, Benson, and Espinoza demonstrate, reliance on a fingerprint system is no guarantee that the government analyst’s ultimate opinion, identifying the defendant, is accurate. In truth, there are a number of potential weaknesses in these systems. The use of a computerized system to identify a culprit is a multi-step process, and there is a possibility of error at every step in the process. For example, there are potential weaknesses at the following, major steps.
Initially Scanning the Image 
Before a computer can alter or enhance an image, the image must first be scanned or otherwise entered into the computer. This step raises an input problem. The available scanners are not accurate enough to perfectly represent the photographic images they are asked to scan. Photographic images are often bent to conform to the shape of a computer screen which is also the shape of most computer printers. “Digital imaging for the most part has a long way to go to meet the quality of film,” said Richard Vorder-Bruegge, an FBI forensic expert who chaired a panel that wrote guidelines for law enforcement use of digital imaging.
In many jurisdictions, Livescan scanners have replaced paper based fingerprint images. Often there are two available settings on Livecan: some detail and more detail. If the operator chooses the first setting, the scanned image can omit a detail that would lead to the positive exclusion of an innocent person such as Mayfield. Selecting the first setting might be convenient, but it sacrifices accuracy.
Indexing the Stored Image
Before the image is stored, it must be linked (indexed) to a specific person. Banking and manufacturing systems frequently automate this step by the use of magnetic ink (MICR) or bar codes. In contrast, almost all criminal justice systems rely on fallible human beings to do the indexing. As the Benson case dramatizes, there can be tragic human error at this step.
Storing the Scanned Image
The indexed, scanned image must be stored for subsequent retrieval. Fingerprints are stored on standard hard drives. However, the hard drives of computer systems are vulnerable; they can be hacked, causing incorrect information to replace correct data. Given the potentially dire consequences of hacking, each data center should be prepared to detect the creation of erroneous information and restore the correct information. However, even some of the largest business firms in the world do not yet have that technical capability. Similarly, many criminal justice data centers currently lack that capability.
Retrieving the Stored Image
Once stored, images can be retrieved and re-enhanced or printed. However, you cannot assume that the printout is identical to the original image. Suppose, for example, that after the storage of the original image, the center purchases a new and “better” printer. Its printout may not only be different than the image the earlier printer would have produced; there is no guarantee that the printout is more accurate than the image which the earlier printer would have yielded.
Manipulation of the Stored Image
As the preceding paragraph indicates, especially when there is a subsequent change in retrieval technology, the change can alter the printout of the originally stored image. Such changes are inadvertent.
However, there is also the possibility of deliberate manipulation of the image. One type of deliberate manipulation is especially problematic — so-called enhancement. For instance, some AFIS workstations claim that they can separate overlapping fingerprints. In order to “enhance” the image in these situations, the computer uses mathematical transforms, formulae which dictate the alteration of the image. Every mathematical transform must be tested to assess its measurements and its scientific validity. The accuracy of the expert’s final opinion depends on the validity of these formulae. Consequently, under Daubert v. Merrell Dow Pharmaceuticals, Inc.,3 the United States Supreme Court’s landmark 1993 decision, the prosecution ought to be required to lay a foundation establishing the validation of each formula.
Digital Fingerprint Standard
Digital images have less detail than the original impressions. Less detail means less exclusionary information. In a criminal investigation, people with freckles and long hair cannot be excluded if the photograph of the perpetrator lacks those essential details. Worldwide, there are already varying fingerprint standards. Some countries require a minimum number of points of similarity. In contrast, in other countries the examiner is permitted to find a match when the details within a small area of the latent correspond to details in the same area of the exemplar.
The advent of digital technology necessitates that we revisit the question of a proper standard for declaring a match. Over 50 percent of exemplars and approximately 25 percent of latent fingerprints are now digital and therefore should be challenged.
Best Hope
There are many common, comforting myths about electronic evidence. However, we must come to grips with the harsh realities: Electronic evidence lacks permanency, it can be modified, it is vulnerable to hackers, the alteration of electronic evidence is almost always visually undetectable and our analysts need new training its use. Sadly, many government agencies that rely heavily on electronic evidence lack the expertise to fully appreciate its limitations.
In the short term, as these three horror stories demonstrate, uncritical reliance on digital information in the justice and homeland security systems creates a grave risk of miscarriage of justice. In the short term, the best insurance for a defendant is thorough discovery and the critical evaluation of the produced evidence. The attorney should file a discovery motion requesting the disclosure of the original image and all computerderived images. After the attorney has obtained all the images, he or she can have them reviewed by an imaging expert to identify any problems such as equipment distortions, electronic enhancements and color.
In the long term, the best hope is the development of professional standards ensuring accuracy and meticulous care at each step in the process culminating in the expert’s final opinion . The Association for Information and Image Management (AIIM) is a worldwide trade organization and an accredited American National Standards developer. At the moment, it is at work in earnest to develop imaging and electronic content guidelines for imaging devices and procedures. The AIIM evidence committee is one of its largest and most active committees. Imaging is such a powerful tool that safeguards are required to protect the public. The safeguards which AIIM developed for the signing and use of credit cards have served the public well for years.
This subject is so complex and of such vital importance that in our judgment, it merits a detailed study by a qualified neutral third party such as the National Academies of Sciences (NAS). In the past, the National Research Council of the NAS has studied such forensic techniques as sound spectrography, DNA typing, polygraphy, and comparative bullet lead analysis. Electronic image processing is a more pervasive technique in our society than any of those forensic methodologies. Mayfield, Benson, and Espinoza demonstrate that we need a much more sophisticated understanding of the limitations of these computerized systems. The NAS could perform a tremendous public service by educating the public and private sectors about the pitfalls of digital techniques.
Digital Discovery
EXEMPLAR SECTION
Please separately answer the appropriate questions for each matched exemplar
1. Identify anatomy matched, i.e., left palm, right thumb etc.?
2. Is this exemplar a rolled, flat or flat group?
3. If this exemplar is a Livescan image, what make/model scanner was used?
4. If this exemplar is paper-based and then computer-scanned, what
make/model scanner was used?
5. Provide the date and origin, e.g. FBI IAFIS, Ohio AFIS, local paper,
local Livescan, Ohio Livescan etc., of this exemplar?
6. If this exemplar is paper-based and copied, what copier, make/model
was used?
7. Explain each computer improvement made, including improvements that
SWIGIT allows.
8. Provide a brief explanation of any incomplete or unknown answers.
EXEMPLAR MATCH SECTION
Please separately answer the appropriate questions for each matched exemplar
1. Was this exemplar matched using a screen, paper or other?
2. Describe other.
3. If this exemplar was screen matched, what is the make and model of
monitor and the video card?
4. If this exemplar was paper matched, what printer wae used?
5. How was the match corroborated: screen, printer output, other.
6. If this exemplar was screen matched, what is the make and model of
monitor and the video card?
7. If this exemplar was paper matched, what printer was used?
9. If AFIS search was performed what was the scope of the search e.g.
national, statewide etc.?
10. Was more than one search performed?
LATENT SECTION
Please separately answer the appropriate questions for each matched latent
1. Identify anatomy matched i.e. left palm, right thumb etc.?
2. Is this latent a lift or an image of a lift?
These questions only apply to an image of a lift
3. Was the image matched on a screen or on paper?
4. How was the original image created: 35 MM film, digital camera,
Polaroid etc.?
5. Provide makes and models of the camera and lens used.
6. Is this a color, grey scale, or a black and white image?
7. If this image is paper-based and then computer-scanned, what
make/model scanner was used?
8. If this latent is paper-based and copied, what make/model copier was
used?
9. Explain each computer improvement made, including improvements that
SWIGIT allows.
10. Provide a brief explanation of incomplete and unknown answers.
|
Notes
We would like to thank Manfred Schenk, Conrad Macina, Harry Whitcomb and Joseph Collins for their technical peer review of this article.
1. http://benson-vs-identix.com/
2.http: //benson-vs-identix.com/
htmfiles/plaintiffs/html#Anchor-Roge-16034
3. 509 U.S. 579 (1993)
© Copyright 2004 Mike Cherry, Edward J. Imwinkelried and Larry Meyer, all rights reserved n
|