Dual-Fidelity Tensor Robust Principal Component Analysis Under Mixed Noise

Abstract

Existing tensor robust principal component analysis (TRPCA) frameworks have achieved remarkable success in recovering low-rank structures from high-dimensional data. However, current methods often rely on a single noise distribution assumption, limiting their effectiveness in complex real-world scenarios and consequently degrading recovery performance. To address this issue, this paper proposes a dual-fidelity tensor robust principal component analysis (DF-TRPCA) framework. Specifically, the proposed method jointly incorporates the sparsity-promoting $\ell _{1}$-norm and a mixture-correntropy-induced (MCI) penalty to characterize sparse corruptions and unstructured non-Gaussian noise, respectively. By integrating these two complementary penalty terms into the TRPCA framework, the proposed method effectively suppresses complex noise while faithfully preserving the intrinsic low-rank structure of the data. Extensive experiments on benchmark datasets demonstrate that the proposed algorithm achieves superior recovery performance.