FIDA: Face Recognition using Descriptive Input Semantics

Generic face recognition systems identify a subject by comparing the subject’s image to images in an existing face database[11]. These systems are very useful in forensics for criminal identification[7] and in security for biometric authentication[8], but are constrained by the availability and quality of subject images[1]. In this paper, we propose a novel system, FIDA, that uses descriptive nonvisual human input of facial features to perform face recognition without the need for a reference image for comparison. FIDA maps images in an existing database to a fourteen dimensional descriptive feature space using softmax regression, and compares input feature descriptions to images in feature space. FIDA clusters database images in feature space using feature-weighted K-Means clustering[4] to offer computational speedup while searching feature space for matching images.