基于张量分解可视化的手势视频分类研究

With the explosive growth of data, higher order data types are becoming more and more popular. Tensor has a natural advantage in representing higher order data. Since hand gesture classification is widely used in various areas, in this paper, we modeled hand gesture video data with three order tensors, and focused on the visualization of hand gesture video tensor by leveraging Higher-Order Singular Value Decomposition (HOSVD). To achieve better classification performance, we introduced canonical angel to measure the distance of tensors. Experimental results conducted on the Cambridge hand gesture database show the promising classification performance with prior knowledge provided by visualization of tensor decomposition.