Novel features selection for gender classification
Jiann-Der Lee, Chun-Yi Lin, Chung‐Hsien Huang · 2013
This paper proposed a novel gender classification system based on selected texture-based features and Support Vector Machine (SVM) classifier. In this study, t-test is applied as a feature selection technique to select significant features. Firstly, we extract texture-based features comprising Local Binary Patterns (LBP) and Histogram of Oriented Gradient (HOG) of face images from FERET face database. Then t-test is employed to determine each feature if it has significant difference between male and female categories. Next, the SVM model is trained with the significant features, which are selected by p-value selection of training samples. Finally, the accuracy of the trained gender classifier is estimated by using testing samples. The experimental results show that with the proposed t-test-based gender classification the number of features is decreased dramatically from 5195 to 1563, a 70% reduction, and the accuracy also shows slight improvement which is from 91.5% to 92.2%.