Real-time embedded age and gender classification in unconstrained video

In this paper, we present a complete framework for video-based age and gender classification which performs accurately on embedded systems in real-time and under unconstrained conditions. We propose a segmental dimensionality reduction technique using Enhanced Discriminant Analysis (EDA) to reduce the memory requirements up to 99.5%. A non-linear Support Vector Machine (SVM) along with a discriminative demographics classification strategy is exploited to improve both accuracy and performance. Also, we introduce novel improvements for face alignment and illumination normalization in unconstrained environments. Our cross-database evaluations demonstrate competitive recognition rates compared to the resource-demanding state-of-the-art approaches.