Unconstrained gender classification by multi-resolution LPQ and SIFT

One of the most critical tasks in building a gender classification is how to describe the human face as a highly discriminative feature vector. To this end, in this paper we introduce a new handcrafted feature extraction method for unconstrained gender classification problem. From one input face image, we generate its smaller version and apply two LPQ operators on both of them. We then combine the obtained LPQ features with the SIFT feature extracted from the given image to constitute a global facial description. In the classification stage, the binary SVM classifier is used for determining the gender of the test images. To evaluate the recognition performance of the proposed methods, we carry out experiments upon two widely used unconstrained face databases Adience and LFW. The results show that our approach attains good classification rates (96.51% and 80.5% on LFW and Adience databases, respectively) and can be comparable with state-of-the-art systems.