NIST Corner: Scientists Automate Key Step in Forensic Fingerprint Analysis
Written by Rich Press   

FOR MOST OF THE LAST CENTURY, fingerprints were considered a nearly infallible method of identification. Recent research has shown, however, that fingerprint examination can produce erroneous results. A 2009 report from the National Academy of Sciences, for instance, found that results, “are not necessarily repeatable from examiner to examiner,” and that even experienced examiners might disagree with their own past conclusions when they re-examine the same prints at a later date. Although errors such as these occur infrequently, when they do, they can lead to innocent people being wrongly accused and criminals remaining free to commit more crimes.

Fingerprints left at a crime scene—so-called latent prints—are often partial, distorted and smudged. Credit: Chugh et al., Hancek/NIST

But scientists have been working to reduce the opportunities for human error. Recently, scientists from the National Institute of Standards and Technology (NIST) and Michigan State University reported that they have developed an algorithm that automates a key step in the fingerprint analysis process. Their research was published in IEEE Transactions on Information Forensics and Security.

“We know that when humans analyze a crime scene fingerprint, the process is inherently subjective,” said Elham Tabassi, an electronics engineer at NIST and a co-author of the study. “By reducing the human subjectivity, we can make fingerprint analysis more reliable and more efficient.”

A Key Decision Point
Computers are already generally able to reliably match high-quality fingerprints to high-quality fingerprints, such as those recorded on a fingerprint scanner. However, matching latent prints from a crime scene often requires human judgement.

“At a crime scene, there’s no one directing the perpetrator on how to leave good prints,” said Anil Jain, a computer scientist at Michigan State University and a co-author of the study. As a result, latent prints are often partial, distorted, and smudged, or left on a confusing background.

A commonly used method for examining latent prints is the ACE-V method, for Analysis, Comparison, Evaluation, and Verification. During the first step of this process (Analysis), the examiner judges how much useful information the print contains. A high-quality print might contain sufficient information to identify an individual. A lower quality print might only be useful for excluding an individual. A still lower quality print may have no investigatory value at all.

“This first Analysis step is standard practice in the forensic community,” said Jain. “This is the step we automated.”

Following the Analysis step, if the print contains sufficient usable information, it can be submitted to an Automated Fingerprint Identification System. The AFIS then searches its database and returns a list of potential matches, which the examiner evaluates to look for a conclusive match.

The Analysis step is critical because, “If you submit a print to AFIS that does not have sufficient information, you’re more likely to get erroneous matches,” Tabassi said. On the other hand, “If you don’t submit a print that actually does have sufficient information, the perpetrator gets off the hook.”

Currently, the process of judging print quality is subjective, and different examiners come to different conclusions. Automating that step makes the results consistent. “That means we will be able to study the errors and find ways to fix them over time,” Tabassi said.

Automating this step also will allow fingerprint examiners to process evidence more efficiently. That will allow them to reduce backlogs, solve crimes more quickly, and spend more time on challenging prints that require more work.

Automatically extracted features of a latent fingerprint: (a) Input latent with manually marked region of interest, (b) ridge flow overlaid on the cropped latent, (c) ridge quality map, and (d) features that can be used as points of comparison, including minutiae (white circles) and core points (green circles). Credit: Chugh et al.

Training the Algorithm
The researchers used machine learning to build their algorithm. Unlike traditional programming in which you write out explicit instructions for a computer to follow, in machine learning you train the computer to recognize patterns by showing it examples.

To get training examples, the researchers had 31 fingerprint experts analyze 100 latent prints each, scoring the quality of each on a scale of 1 to 5. Those prints and their scores were used to train the algorithm to determine how much information a latent print contains.

After training was complete, researchers tested the performance of the algorithm by having it score a new series of latent prints. They then submitted those scored prints to AFIS software connected to a database of over 250,000 rolled prints. All the latent prints had a match in that database, and they asked AFIS to find it.

This testing scenario was different from real casework because, in this test, the researchers knew the correct match for each latent print. If the scoring algorithm worked correctly, then the ability of AFIS to find that correct match should correlate with the quality score. In other words, prints scored as low-quality should be more likely to produce erroneous results—that’s why it’s so important to not inadvertently submit low- quality prints to AFIS in real casework—and prints scored as high-quality should be more likely to produce the correct match.

Based on this metric, the scoring algorithm performed slightly better than the average of the human examiners involved in the study.

What made this breakthrough possible, besides recent advances in machine learning and computer vision, was the availability of a large dataset of latent prints. Machine-learning algorithms need large datasets for training and testing, and until now, large datasets of latent fingerprints have not been available to researchers, largely due to privacy concerns. In this case, the Michigan State Police provided the researchers with the testing dataset, after having first stripped the data of all identifying information.

The next step for the researchers is to use an even larger dataset. This will allow them to improve the algorithm’s performance and more accurately measure its error rate.

“We’ve run our algorithm against a database of 250,000 prints, but we need to run it against millions,” Tabassi said. “An algorithm like this has to be extremely reliable, because lives and liberty are at stake.”

About the Author
Rich Press is a writer with NIST.

T. Chugh, K. Cao, J. Zhou, E. Tabassi and A. K. Jain. Latent Fingerprint Value Prediction: Crowd-based Learning. IEEE Transactions on Information Forensics and Security. Published online 28 June 2017. DOI: 10.1109/TIFS.2017.2721099

This article appeared in the Winter 2017 issue of Evidence Technology Magazine.



< Prev   Next >

Lifting Latent Fingerprints from Difficult Surfaces

ALMOST ANYONE can find, process, and lift a latent print that happens to be in a logical and obvious place like a door handle, a beer can, or a butcher knife. But sometimes, a latent print is not just sitting there in a logical and obvious place. Sometimes, you have to use your imagination to find the print and your skills to lift it.