Hierarchical Fusion for Gender Recognition Based on Hand Images

Manman Zhang, Yuchun Fang, Yifan Li

Abstract: Human hand features contain rich gender information. In this paper, we propose an approach to boost the performance of a given gender recognizer through the hierarchical fusion of human hand subspace features, texture features, and geometric features. We first use Eigenhand to extract subspace features. Hand texture features are obtained by applying local binary mode histograms. Then we get the geometric features by calculating the length ratio of fingers other than the thumb. We call this method as Geometric Gender Descriptor of Hand Images (GGD-H). Based on these, we use the serial strategy to fuse texture features and geometric features in the feature-level and then fuse its classification result with the subspace feature in the decision-level. The fused vectors are used as the final feature vectors to feed the support vector machine for our gender recognition task. Through this method, we can obtain gender difference features from multiple directions, thereby enhancing the perception ability of the same human hand in different scenes and the robustness of feature expression. The final experimental results show that the method proposed in this paper achieves a recognition rate of 0.988 on 11K Hands, exceeding the common hand gender recognition scheme.

External IDs:dblp:conf/prcv/ZhangFL20