A New Vision Two-Class Classification Method Based on Tensor Technology

摘要

In this paper, we introduce the recent progress in the vision classification based on tensor technology, especially concentrating on Support Tensor Machines (STM) and Support Tucker Machines (STuM). A model is proposed to select the dimension of Tucker decomposition in STuM, and it can be transformed into an optimization problem with low-rank constraint. After analysis and transformation of the problem, we have found that its penalty problem can be solved by the proximal subgradient method.