A hybrid approach to gender classification from face images

Recently, gender classification from face images has attracted a great deal of attention. It can be useful in many places. In this paper, a novel hybrid face coding method by fusing appearance features and geometry features is presented. We choose Haar wavelets to represent the appearance features and use AdaBoost algorithm to select stronger features. Geometry features are regarded as apriori knowledge to help improve the classification performance. In this work, active appearance model (AAM) locates 83 landmarks, Thus we can get 3403 geometry features, from which 10 most significant features are picked, normalized and fused with the appearance features. Experimental results show the effectiveness and robustness of the proposed approach regarding expression, illumination and pose variation in some degree.