The Immutable Laws of Software Development

The lessons and best practices that have become required operating procedure in software development groups can often be applied outside the immediate field of software engineering. This article details a groundbreaking new, multi-year, large-scale biometric dataset that is designed to improve the accuracy and robustness of iris recognition algorithms. We identify several challenges associated with this collection effort and demonstrate how the application of software best practices was able to overcome these obstacles. We believe this list of recommendations represents the current best practices for large scale, long-term biometric collections. Collecting Large Biometric Datasets A Case Study in Applying Software Best Practices percentage of trials where a single person appears to not match their own biometric sample, usually requiring the individual to resubmit their test sample. High-quality commercial iris systems can maintain a FMR of one in one million matches while sustaining an FNMR of one in every one thousand attempts [2]. These extremely accurate metrics make iris biometrics one of the few that are appropriate for fully automated populationscale identification programs. Table 1 details some of the large national programs initiated in the last decade. In 2007, the United States military also began utilizing mobile iris biometric technologies. These aptly named devices, known as the Handheld Interagency Identity Detection Equipment (HIIDE) and Secure Electronic Enrollment Kit (SEEK) were deployed to battlefields in both Iraq and Afghanistan to assist with base access, detainee management, local population screening, and special operations missions. By 2009, the Biometrics Identity Management Agency, which executes biometrics initiatives for the DoD, had collected more than 7.5 million iris images in the field [3].