JNET & PennDOT Facial Recognition Integration CATEGORY: DATA, INFORMATION & KNOWLEDGE MANAGEMENT Contact: David Naisby JNET Executive Director 5 Technology Park Harrisburg, PA 17110 (717) 214-7461 dnaisby@pa.gov Commonwealth of Pennsylvania Project Initiated: September 2012 Project Completed: December 2012
Executive Summary The Pennsylvania Justice Network (JNET) is an integrated portal that provides authorized users with access to public safety and criminal justice information from federal, state and local sources. Over 39,000 municipal, county, state and federal justice professionals use JNET to access critical information and securely conduct investigations. JNET provides role-based access to over 35 distinct applications. In addition, JNET uses a service oriented architecture (SOA) model to broker the exchange of over 600 million messages annually between business partners. The JNET infrastructure connects all 67 Pennsylvania counties, 38 state agencies, and 37 federal law enforcement agencies. JNET has deployed ground-breaking technology in the field of facial recognition. The JNET Facial Recognition System (JFRS) allows law enforcement investigators to compare images from a variety of sources, including surveillance or security video and social media sites, against a statewide criminal database containing 3.5 million photographs. A major limitation of the legacy JFRS system was that it only allowed comparison of images for previously arrested criminal offenders. At the request of the law enforcement community, JNET partnered with the Pennsylvania Department of Transportation (PennDOT) to enable JFRS searches against the statewide repository of 36 million driver s license and identification photo images. Leveraging JNET s accessibility and state-of-the-art facial recognition system with PennDOT s vast photo database has provided immediate and exponential value for all law enforcement professionals in Pennsylvania. Through this solution, investigators now have the ability to submit a photograph or image into JFRS, and compare it against approximately 40 million images using three unique search algorithms. This project was a collaborative effort between the Office of Administration, JNET, PennDOT, and the Pennsylvania Chiefs of Police Association. This effort has reduced costs, saved time, and strengthened public safety across the state.
3. Business Problem and Solution Description The ability to identify an unknown suspect or witness is paramount in the prosecution and prevention of crime. To support this ability, JNET deployed a facial recognition system (JFRS) that enables trained and authorized investigators to upload images for comparison against 3.5 million criminal booking photographs. JFRS provides a host of imaging tools that allow photographs to be cropped, rotated or manipulated to assist with comparison. Furthermore, JFRS employs two unique search algorithm technologies. Photographs submitted into JFRS by investigators are automatically analyzed by both search algorithms, ensuring that law enforcement benefits from multiple technologies. With JFRS deployed on JNET, the system can be made available to any law enforcement agency in Pennsylvania. Law enforcement professionals from more than 500 agencies have completed training and have access to JFRS. JFRS is able to identify similarities between the unknown person or suspect using the unique personal measurements of an individual s facial features. These measurements are compared to offender photographs captured at the time of criminal booking within the statewide database known as the Commonwealth Photo Imaging Network (CPIN). During the booking process, a photograph is captured in CPIN and a facial plate is automatically created from the individual s picture. Photographs stored in CPIN are made available to law enforcement through a web-based application on JNET called WebCPIN. Pennsylvania has over 3.5 million criminal booking photographs for comparison in JFRS. Unfortunately, this represents only a fraction of the approximately 13 million residents in the state. Without a larger database for comparison, law enforcement was limited in its ability to investigate unknown criminal suspects or suspects who have never been arrested or booked in Pennsylvania. The Pennsylvania Department of Transportation (PennDOT), which is responsible for driver s license and identification cards issued in Pennsylvania, maintains a database of over 36 million photographs. PennDOT, concerned with the issuance of duplicate or fraudulent identifications, maintains its own internal facial recognition solution. However, without the network infrastructure and connectivity of JNET, the PennDOT facial recognition solution was only available to limited users in the Pennsylvania State Police (PSP) and Pennsylvania Office of Attorney General (AG).
The challenge facing Pennsylvania law enforcement was how to integrate two distinct facial recognition systems with contrasting strengths. JNET had an award-winning dual algorithm technology that was readily available to all law enforcement in the commonwealth but it could only match against suspects with a prior arrest history. PennDOT had a facial recognition solution with limited availability but could match against Pennsylvania s extensive driver s license and identification card photo database. Recognizing the value of providing driver s license and identification images to JFRS investigators, the JNET Steering Committee charged JNET and PennDOT with architecting a solution that would augment law enforcement s investigative abilities while respecting policy, privacy and fiscal considerations. A workgroup was formed between the two agencies that included technical, business and vendor representation. The group immediately went to work assessing agency system capabilities, drafting requirements and - most importantly reviewing privacy and security policy considerations. Early in the process, it became clear that federal and state policies limited the ability to share photographs between the two systems. This meant that faceplates from one system could not be copied or shared with the other system. To ensure that these policy requirements were met, the team decided to use JNET s SOA infrastructure to develop a set of web services that would facilitate the loosely coupled integration of the systems. This approach proved to be simple, efficient, cost-effective and secure. When uploading the photograph of an unknown suspect into JFRS, the system automatically invokes a web service to PennDOT. This web service submits the facial plates of the unknown suspect or probe image to PennDOT for comparison against its license and identification photo database. As potential matches are identified, PennDOT then invokes a web service to return the potential matching results for use in JFRS. In the meantime, JFRS processes the suspect probe image against previously known criminal photos using its two internal search algorithms.
As a result, JFRS now leverages three unique facial recognition search algorithms. More importantly, investigators now have more than 10 times the potential images to compare against when investigating crimes. The integration solution had minimal impact to existing JFRS users. By using web services, JNET was able to integrate results from PennDOT directly into the JFRS interface. There was no need for users to learn a new system and trained investigators were simply provided a third set of image results to view. Training documentation was updated to reflect the new PennDOT results and was made available to the existing JFRS users. The integration of the two systems was so seamless that JNET has not received a single support call for the new interface. The entire project took less than six months to conceptualize, design and implement. The core project team included subject matter experts and quality assurance staff from JNET, development staff from each of the facial recognition providers and project management from PennDOT. Policy, application support, and management representatives from both PennDOT and JNET participated as needed. PennDOT images were made available to JFRS users on December 18, 2012. This project cost the commonwealth $75,000 to develop, test and implement, and was made available at no cost to JNET s user community. 4. Significance This project benefits law enforcement by providing a single facial recognition solution that uses three distinct search algorithms to query against two very large repositories that collectively contain nearly 40 million images. With JFRS available through JNET, this enterprise solution is available at no cost to any municipal, county, state or federal law enforcement agency in the commonwealth. Pennsylvania is not aware of any other facial recognition platform in the United States that uses this technology set, or one that is available to such a large and diverse group of users. The use of web services to bridge the gap between PennDOT and JFRS allowed both agencies to maintain the ownership and control of their data while meeting federal and state policies designed to protect the privacy and security of the public.
Developing consensus and buy-in from stakeholders to complete this effort was relatively easy. The project clearly aligned with several administration and JNET goals: Develop and maintain strategic partnerships. Facilitate efficient government operations. Provide a contemporary gateway for the delivery of public safety and criminal justice data among municipal, county, state, and federal consumers. Provide extraordinary customer service to the public, agency business partners and stakeholders, and the user community. Finally, any effort that increases the effectiveness and abilities of criminal investigators has a direct positive impact on public safety. Furthermore, improved public safety is a benefit that can be enjoyed and appreciated by all members of the public. 5. Benefits of the Project The most immediate benefit of this project is in the area of public safety. With a minimal investment by the commonwealth, criminal investigators from any jurisdiction in Pennsylvania now have access to 10 times as many potential matches as they did before this project was implemented. Feedback from users of the system has been unanimously positive with over 400 new users trained to access the system in 2013 alone. The value of adding an additional search algorithm with millions of records with no need for new training or additional costs cannot be overstated. Recently, users have reported using the JFRS application on mobile devices to identify unknown individuals during traffic stops; a direct benefit of providing drivers license photographs through JFRS. The project is an example of government at its best: implementing solutions using industry standards and best practices for the benefit of all citizens: Strong governance allowed the agencies the authority to pursue a solution without delay or unnecessary roadblocks. Use of the JNET Private Key Infrastructure (PKI) and role-based identity management solution ensured that data and services were accessed and shared securely. As an enterprise-wide shared service, the PennDOT integration into JFRS provides immediate, no-cost benefits to users.
In addition, through the adoption of interoperable and standards-based service development practices such as Global Reference Architecture (GRA) and the National Information Exchange Model (NIEM), the commonwealth has ensured that the services developed for this interface can be reused and shared when the need arises. In summary, Pennsylvania has developed an enterprise solution that merged the functionality and data sets from two disparate facial recognition systems at minimal costs with no impact to users. This dramatically advanced public safety in the commonwealth by augmenting law enforcement s ability to search and identify unknown suspects, regardless of agency size, technology disposition or budget. All of this was accomplished without impacting the integrity, ownership or control of the agency data or applications driving the solution while protecting citizen s privacy.