Enhancing Gender Classification by Combining 3D and 2D Face Modalities

Baiqiang Xia (1) , Boulbaba Ben Amor (1) , Huang Di (2) , Daoudi Mohamed (1) , Wang Yunhong (2) , Drira Hassen (1)

1 LIFL - FOX MIIRE
2 School of Mathematics and Systems Science

Shape and texture provide different modalities in face-based gender classification. Although extensive works have been reported in the literature, the majority of them are in the scope of shape or texture modality individually. Among them, only a few concern their combination, and to the best of our knowledge, no work considers the combination with the 3D face surface. In our work, we investigate the combination of shape and texture modalities for gender classification, with both the combination of range images and gray images, and the combination of 3D meshes and gray images. In 10-fold subject-independent cross-validation with Random Forest on the FRGC-2.0 dataset, we achieved a correct gender classification rate of 93.27%± 5.16, which outperforms each individual modality and is comparable to the state-of-the-art. Results confirm that shape and texture modalities are complementary, and their combination enhances the performance of face-based gender classification.

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]

Baiqiang Xia, Boulbaba Ben Amor, Huang Di, Daoudi Mohamed, Wang Yunhong, et al.. Enhancing Gender Classification by Combining 3D and 2D Face Modalities. 21th European Signal Processing Conference (EUSIPCO), Sep 2013, Morocco. pp.1-6. ⟨hal-00829884⟩