Weighted low-rank t-SVD under general mixture noise distribution

To enhance the adaptability of low-rank tensor singular value decomposition (LRTSVD) to various superposition noises, we introduce an innovative LRTSVD model. This model considers noise as a mixture of exponential power (MoEP) distributions and further extends into a penalized MoEP (PMoEP) model by integrating the penalized likelihood approach with MoEP distributions. The model parameters are estimated using the Expectation-Maximization (EM) framework, complemented by an innovative algorithm tailored for low-rank tensor-SVD. Comprehensive experiments on multidimensional simulated datasets have demonstrated the PMoEP-LRTSVD algorithm's superiority over existing competitors.