Hand gesture recognition using topological features

Hand Gestures Recognition (HGR) is one of the main areas of research for Human Computer Interaction applications. Most existing approaches are based on local or geometrical properties of pixels. Still, there are some serious challenges on HGR methods such as sensitivity to rotation, scale, illumination, perturbation, and occlusion. In this paper, we study HGR from graph viewpoints. We introduce a set of meaningful shape features based on a graph constructed by Growing Neural Gas (GNG) algorithm. These features are constructed from topological properties of this graph. Graph properties in conserving topological features improve stability against different deformations, scale, and noise. We evaluate our method on NTU Hand Digits dataset with state-of-the-art methods. We also prepared a comprehensive dataset (SBU-1) for different hand gestures containing 2170 images. This dataset includes many possible deformations and variations and some articulations. Most of the existing datasets don’t capture these variations. We show the robustness of the algorithm to scale, rotation and noise, while preserving similar recognition rate in comparison with the state-of-the-art results.