2/18/11

Gender Classification of Face Images

Gender recognition by face is one of the actual problems of computer vision. Gender recognition could be useful in the number of applications for biometric authentication, hightech surveillance and security systems, criminology, automatic psychophysiologic inspection, augmented reality etc. Also the applicability of gender recognition is growing in a such areas as social science, statistics and marketing research. Also there are a lot of applications (especially in social nets) based on different face recognition algorithms (including sex classification) for entertainment of users. Thats why sex recognition by face is of interest by computer vision scientists during last two decades.

Gender recognition can be regarded as classification problem of detected faces into classes (males & females). The gender recognition task is being investigated from the beginning of 90-th of XX-th century. The best of the state-of-the-art results reported in scientific papers are about 95% accuracy. After testing of several the most promising approaches we succeeded in achieving 96% for male and 95% for female faces on FERET face image database. It was achieved on LBP (Local Binary Patterns) features classified by SVM (Support Vector Machine) with RBF (Radial Basis Function) kernel function.

1/24/11

Face Tracking using Face Identification and Pattern Tracking

Face tracking is an important issue in many computer vision problems, particularly in the video surveillance area. Face tracking extends face detection which takes place at the still images to video sequences where additional spatio-temporal information can be used. So, using face tracking allows to obtain more information about faces in the video.
The goal of face tracking thus consists in aggregation of single-frame detection results into a collection of face tracks. For example, we can count number of different faces appeared in the camera area during all observation period and register time intervals of the presence of each face, obtain number of faces presented in the frame at the current moment and follow their positions.
video

10/12/10

Video Surveillance with ZoneMinder and Computer Vision Plugins

ZoneMinder is a powerful video surveillance software for Linux intended for use in single or multi-camera video security systems. ZoneMinder supports wide range of USB and IP cameras and allows flexible configuration for any hardware.
But its main feature is alarms invoking and video recording based on events occurrence (for example, if motion detected within observed area). Such events detection based on video analysis opens up great possibilities for intellectual video surveillance. 
The only one video analysis feature provided by ZoneMinder at the moment is quite simple motion detection algorithm. ZoneMiner in its current state hasn't plugins support for performing additional video analysis for events detection.