Automatic Gender Discrimination from Video Sequences of Human Gait

Automatic gender identification plays an important role in identification of a person. We have presented a study and analysis of gender classification based on gait. In this work gait of a person is represented by simple binary moment features of seven regions such as head/shoulder region, front of torso, back of torso, front thigh, back thigh, front calf/foot and back calf/foot. These features are computed from parameter values of ellipses that fit body parts enclosed by different regions. Then the extracted features are used for training and testing different pattern classifiers like kNN (k-Nearest Neighbor) and SVM (Support Vector Machine) to classify the gender. Apart from accuracy, other measures more appropriate for imbalanced problems are also considered in this paper. We analyzed the performance of SVM with various kernel types and kNN with various distance measures and k (number of neighbors)value. Experimental results show that SVM classifier with linear kernel gives better results when compared to kNN classifier. The classification results are more reliable than those reported in previous papers. The proposed system is evaluated using side view videos of CASIA dataset B.