Context-aware face recognition

Abstract

Current face recognition systems make extensive use of class-wise disparate features drawn from faces in order to achieve state-of-the-art performances. However, there exists a multitude of information in a given image that is filtered out due to the current systems being restricted to focusing entirely on a face. This information prominently includes co-occurrence of certain individuals together, an individual’s gait or frequently visited locations. We have designed a system to utilize such context-based information as an additional biometric score which is fused with the score obtained by state-of-the-art recognition systems to theoretically improve the performance of our vanilla recognition systems.