A robust facial representation is an essential component for gender classification. This paper introduces a new local feature, Local Gradient Increasing Pattern (LGIP), which expresses the local intensity increasing trend. A LGIP feature is to encode intensity increasing trends in 8 orientations at each pixel using signs of directional gradient responses, and overall increasing trend is assigned with a decimal label. A facial image is partitioned into overlapping regions from which LGIP histograms are obtained and concatenated into a single feature vector. Gender classification is carried out using SVM classifier based on the LGIP-based facial descriptor. We investigate the influence to recognition rates by two factors, image resolution and person-dependent/independent condition. Experiments are performed on two replicable image sets from CAS-PEAL and FERET databases, and the results show that our method achieves better performance than many other methods. © 2012 Springer-Verlag.
CITATION STYLE
Zhou, L. B., & Wang, H. (2012). Local Gradient Increasing Pattern (LGIP) for facial representation and gender recognition. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 7325 LNCS, pp. 46–53). https://doi.org/10.1007/978-3-642-31298-4_6
Mendeley helps you to discover research relevant for your work.