I-Corps: Exploiting matching score distributions to improve biometric recognition

University Of Houston, Houston TX

Investigators

Abstract

Whether we want to unlock our cell phones or pay with our credit cards, we need to verify our identity. Biometric technologies can help accomplish this task easier, faster and in a more secure manner. For example, Apple's Touch ID system allows users to unlock their phones and make payments using their fingerprint. However, conditions such as noise in the background (in the case of using speech) or bad illumination (in the case of using facial images) can degrade the accuracy of these systems. This problem is more pronounced for challenging recognition tasks such as surveillance, access control, and personalized customer service. Acuity?s latest report estimates that $20 billion in annual revenues will be generated in the biometrics sector from direct purchase and software development fees by the end of 2020. The proposed add-on software developed by this I-Corps team can improve the system's decision-making process regardless of the biometric train used (e.g., face, speech, or iris). Potential customers can either add the proposed software to their biometric system or use a cloud-based app. Recognition algorithms compare biometric data (e.g., speech segments or facial images) to produce scores that reflect how similar are these samples. However, poor acquisition conditions degrade the quality of the obtained scores. This problem affects all biometric systems regardless of the biometric trait used. Score normalization methods transform scores into a domain that reflects similarity more accurately. This team has developed a framework that describes how to employ existing methods (and those to be invented) more effectively. Specifically, the proposed invention is an algorithm that partitions a set of scores into subsets and then normalizes the scores of each subset independently. Participating in the NSF I-Corps program will allow the team to test its hypotheses and improve its chances of converting companies to clients. At the end of the program, the team intends to present a refined business model canvas and demonstrate its minimum viable product.