In movies and television shows, forensic scientists can pick up a partial fingerprint using a satellite camera 500 miles in the sky — and use it to identify the bad guys. All they have to do is repeatedly hit the ENHANCE button, and the problem is solved. In real life, not so much.
In fact, fingerprints already on a fingerprint card can be almost impossible to use — because the print is smeared or incomplete, because the prints are not aligned on the card, because the paper they’re printed on looks more like used toilet paper than cardstock, or even because the prints are not in the right boxes. Even if the print is perfect, getting paper fingerprint cards into a readable, searchable electronic format is an issue – especially at scale. It’s probably true to say that NCIS wishes technological reality were more like the TV show, “NCIS.”
Novetta’s RISER Project has brought Hollywood and reality closer together by making it possible to ingest and read foreign fingerprint cards in a common digital format. An integrated suite, RISER was developed in collaboration with NCIS, CTTSO, OUSD A&S, and SAF/CDM over a four year period. Its components – RISER1, DBITE, RISER2, and RISER CAPTURE – work together to make the bulk collection of foreign biometrics fast, efficient, and cost-effective.
Using RISER, the United States Government (USG) has saved thousands of hours of labor and decades of ingestion time – and identified many persons of interest in one use alone.
The recurring theme in RISER’s development has been to make the impossible possible.
Before the advent of RISER, no one in the intelligence community was doing bulk ingestion of paper fingerprint cards because it was too labor-intensive and expensive to convert them into a usable digital format. Foreign biometrics are frequently available only in paper form, i.e. fingerprint cards. Although the biometric and biographic information on these cards is valuable for USG interests, fingerprint cards effectively are unusable at scale unless converted into a standard USG digital biometric format, EBTS.
RISER bridges the gap; it converts paper fingerprint cards gathered from nations around the world into USG’s digital EBTS format. Those cards vary widely in format, quality, materials, and layout. RISER translates its biometric inputs into a standard format, which enables timely access to full biometric and biographic information in now-digital fingerprint records by USG matching systems such as an Automated Biometric Identification System (ABIS).
RISER Performance and Results
RISER1 reduced the labor required to ingest a particular class of previously-collected fingerprint cards by a factor of 100x or more. By lowering the required effort, RISER1 made it possible to add these biometrics to USG’s matching systems.
Similarly, RISER DBITE enabled NCIS to clear out a ten-year backlog of a class of foreign biometrics. In doing so, NCIS was able to identify persons of interest and refer these findings to USG partner nations for further action.
It’s not just NCIS. FBI CJIS identified operationally relevant results in their biometric data by processing a backlog of foreign fingerprint cards with an early release of RISER2 obtained via FBI’s partnership with NCIS. Armed with RISER2, the FBI saved more than 4,400 labor hours processing a single data set.
NCIS’s recent results with RISER are similarly positive. Before RISER, an NCIS card digitization mission to Ghana for ~120,000 cards required 21 personnel approximately 3,525 hours over 21 days. A similar mission to Guatemala using RISER required just 7 personnel approximately 48 hours to digitize ~57,000 cards with 90% of the data ready for submission into USG databases 2 days after scanning was completed.
Under the Hood – RISER Technology
The technologies found in RISER’s components reflect the dynamic operational priorities of its users:
- In RISER1, the task of realigning previously collected but misaligned fingerprint data was solved using a combination of the open-source NIST NBIS toolkit for reading and writing EBTS files, classic image processing techniques (blob detection, noise removal, clipping), and custom code to segment and realign fingerprints of highly variable image quality.
- In RISER DBITE, the biometric input data appeared in a foreign PDF-based format, a structured PDF in which the biometrics and biographics of a fingerprint card were embedded in the PDF as separate objects. Using our background in digital forensics, we parsed the PDFs into their constituent objects and created clean output EBTS files with both biometrics and biographical metadata populated.
- In RISER2, the set of possible input biometrics were expanded to include full-page images from a flatbed scanner, and full-page images in an unstructured PDF (a PDF file containing one logical object – the page image – per page). Extracting valuable data from such a card image required three stages:
- A neural net model – an image classifier – was implemented with the YOLO DARKNET deep learning framework and trained to identify the orientation of the input page image. The page orientation was corrected so that the working page image would be “right side up” for subsequent processing stages.
- A second neural net – an object detector for biometric segmentation – was implemented with YOLO DARKNET and trained to locate biometrics of interest (fingerprints, palm prints, face images) in the input card image.
- An OCR engine (open source Tesseract) was applied to identify and extract biographical text data in the card image, and to help identify the type of fingerprint card being processed.
- RISER CAPTURE’s task was to supercharge the biometrics digitization process itself. To do this, RISER CAPTURE used a 40MP digital camera mounted on a custom capture rig with integrated lighting. The RISER capture rig replaced a slow and bulky flatbed scanner used for scanning paper fingerprint cards into digital form; a scan of a fingerprint card could take up to 2 minutes to perform with a flatbed scanner, but with RISER CAPTURE this digitization process took about 5 seconds per page. RISER CAPTURE produces high-quality 500DPI or 1000DPI images of fingerprint cards or other biometrics which are converted to EBTS format via RISER2’s neural net-enabled processing logic.
The original 2016 RISER concept called for an integrated forward-deployed appliance – a Rapid Ingest Scalable Entity Router – that would push USG’s biometric matching capabilities to the edge of the Defense Intelligence Enterprise. Think of this as “biometrics matching at the pointy end of the spear.”
Although the RISER name stuck, the concept of operations for RISER swiftly evolved to address more immediate challenges faced by NCIS. They had a tranche of previously collected and digitized foreign fingerprint cards, but the resulting digital records had fingerprint alignment and other quality problems requiring extensive manual effort to bring them up to a high enough quality to ingest into USG matching systems. The volume of collected cards and the amount of time it would take to manually correct them made it uneconomic for NCIS to fix these cards and submit them to ABIS. By uneconomic we mean the cost and time required meant that those cards would never make it into ABIS.
There is where Novetta stepped in to address the problem. The first component of RISER (“RISER1”) implemented an automated and cleanup procedure that enabled the ingestion of 30,000 foreign fingerprint cards previously collected by NCIS. The RISER1 processing run took five days to process the 30,000 cards and had a 90% success rate.
The second component of RISER (“DBITE” or “Digital Biometric Image and Text Extractor”) focused on adding a new input modality to RISER: a PDF-based foreign digital biometrics format in which fingerprints, palm, and face images were embedded as discrete objects in the source PDF. This effort enabled the efficient conversion of approximately 90,000 fingerprint cards of foreign data. This data, a 10-year backlog of NCIS’s collected fingerprint cards of this type, was processed and made ready for ABIS by DBITE in a 36 hour run on a single workstation-class system.
The third component of RISER (“RISER2”) addressed the conversion of raw full-page image scans of fingerprint cards into EBTS format. This effort used a deep learning object detector based on YOLO DARKNET, together with the open source Tesseract OCR engine, to produce EBTS files enriched with full biometrics and biographical metadata.
The fourth and most recent RISER component (“RISER CAPTURE”) included a 40MP digital camera-based biometric capture station. It streamlined the physical scanning process of fingerprint cards into digital image form, reducing the per-page scanning effort from 2 minutes to 5 seconds per card. For bulk card collection missions, this greatly reduces the number of NCIS personnel required and time-on-target spent downrange collecting foreign fingerprint cards.
In August 2020, NCIS won the FedID 2020 Best Technical Advancement Award for its RISER-powered foreign card scanning missions, which have contributed to the success of the federal identity community.