Face-based gender recognition using compressive sensing

In this paper, face images are analyzed in frequency domain and then classified based on the gender of the human subjects appearing in the image. Different images of the same gender are considered as an ensemble of inter-correlated signals and changes due to variation in faces are sparse with respect to the whole image. We exploit this sparsity using compressive sensing, which enables us to grossly represent images of a given gender by two kinds of features: one that represents the common features of the face and the other that denotes the different faces in all training samples. The 1^st and 2^nd features are combined as a cascaded filter for robust gender recognition. The performance of recognition rate is up to 97.64% in our YZUS database and 90.83% in the SUMS benchmark database. Therefore, experiments show the efficacy of our proposed approach.