Boosting Local Gabor Binary Patterns for Gender Recognition

Gender recognition of face images is one of the fundamental face analysis tasks with multiple applications. This paper presents a novel method of gender recognition by using boosting local Gabor binary patterns (LGBP). Local Binary Pattern (LBP) is an effective method for texture description and has been used in a lot of applications. LBP captures the local appearance details while Gabor wavelets encode facial information over a broader range of scales. In order to acquire a better performance, we combine these two complementary methods. Since the feature sets are high dimensional and not all bins in the LGBP histogram are necessary to contain discriminative information for gender recognition, we propose to use Adaboost to select the discriminative features. Promising results are obtained by applying Support Vector Machine (SVM) with the boosted LGBP features.