Gender Classification Based on Feature Selection Using Genetic Algorithms

In this paper we propose a new method to classify gender from face images. In face eigenspace, certain eigenfaces encode more gender information than others. We propose a framework using genetic algorithms (GA) for gender classification. First, GAs are employed to select gender-related eigenfaces from the whole eigenface space and Support Vector Machines (SVM) are used to classify the projection coefficients. Then, GAs are used to select the female eigenfaces and the male eigenfaces from the gender eigenface subset, using SVMs to classify the error between the reconstructed images and average gender images. Finally, in order to fuse the outputs of SVMs from the previous two stages and get the classification decision, the sum rule is employed. Experimental results show that the accuracy of gender classification is 95.1%.