Soft biometric gender classification using face for real time surveillance in cross dataset environment

Gender classification is a challenging task in surveillance videos due to their relatively low solution, uncontrolled environment and viewing angles of an object. It has potential applications as well in visual surveillance and human-computer interaction systems. While a lot of work have considered still face images for soft biometrics recognition and applied still image-based methods, recent developments indicated that excellent results can be obtain on moving faces using texture-based spatiotemporal representations to describe and analyze faces in videos. This paper investigates the combination of facial appearance and motion for face analysis in videos. We proposed an approach for gender classification in spatiotemporal environment from videos by using huge set of training features derived from rich collection of various datasets. We tested our system with several publicly available videos, which have been taken in un-controlled environment in terms of background, light, expression, motion, angle and appearance. We also tested our system with several self recorded surveillance videos. Our extensive cross dataset experimental analysis clearly assessed the promising performance of our system for gender classification using faces in videos. Another novel part of our current research negates the recent theory based on experimental results which claimed that the combination of motion and appearance is only useful for gender analysis of familiar faces.