A Two-Stage Approach to Robust Tensor Decomposition

The rapid advance in sensor technology and computing systems has lead to the increase in the availability of multidimensional (tensor) data. Tensor data analysis have witnessed increasing applications in machine learning, data mining and computer vision. Traditional tensor decomposition methods such as Tucker decomposition and PARAFAC/CP decomposition aim to factorize the input tensor into a number of low-rank factors. However, they are prone to gross error that may occur due to illumination, occlusion or salt and pepper noise encountered in practical applications. For this purpose, higher order robust PCA (HoRPCA) and other robust tensor decomposition (RTD) methods have been proposed. These methods still have some limitations including sensitivity to non-sparse noise and high computational complexity. In this paper, we introduce a two-stage approach that combines HoRPCA with Higher Order SVD (HoSVD) to address these challenges.