Graph Spectral Domain Features for Static Hand Gesture Recognition

The graph spectral processing is gaining increasing interest in the computer vision society because of its ability to characterize the shape. However, the graph spectral methods are usually high computational cost and one solution to simplify the problem is to automatically divide the graph into several sub-graphs. Therefore, we utilize a graph spectral domain feature representation based on the shape silhouette and we introduce a fully automatic divisive hierarchical clustering method based on the shape skeleton for static hand gesture recognition. In particular, we establish the ability of the Fiedler vector for partitioning 3D shapes. Several rules are applied to achieve a stable graph segmentation. The generated sub-graphs are used for matching purposes. Supporting results based on several datasets demonstrate the performance of the proposed method compared to the state-of-the-art methods by increment 0.3% and 3.8% for two datasets.