Face and gender classification in crowd video

Research in face and gender recognition under constrained environment has achieved an acceptable level of performance. There have been advancements in face and gender recognition in unconstrained environment, however, there is significant scope of improvement in surveillance domain. Face and gender recognition in such a setting poses a set of challenges including unreliable face detection, multiple subjects performing different actions, low resolution, and sensor interoperability. Existing video face databases contain one subject in a video sequence. However, real world video sequences are more challenging and generally contain more than one person in a video. This thesis provide the annotated crowd video face database with more than 200 videos pertaining to more than 100 individuals, along with face landmark information and gender annotation to encourage research in this important problem. We provide two distinct use-case scenarios, define their experimental protocols, and report baseline verification results existing on two face recognition systems, OpenBR and FaceVACS. Gender classification is also performed on this database and the results are reported using OpenBR along with a combination of different feature extractors with SVM classification. The results show that both the baseline results do not yield more than 0.16 genuine accept rate at 0.01 false accept rate. A software package is also developed to help researchers evaluate their systems using the defined protocols.