Direct ICA on data tensor via random matrix modeling

Independent Component Analysis (ICA) is a fundamental method for Blind Source Separation (BSS). Classical ICA takes data matrix input formed by vector data. This paper focuses on BSS with third-order tensor data, such as 2D images. Two approaches exist this problem. The first approach reshapes each into to apply classical ICA, structural information lost. second unfolds along different modes perform mode-wise, which partially preserves structures but has strong or ill assumptions. proposes third via RAndom Matrix (RAMICA) modeling. RAMICA works directly, without vectorization unfolding, and row column under more general We develop the model, algorithm, related theories defining new statistics random matrices procedures whitening independent component estimation. study identifiability, higher-order extension, relationships existing methods. Experiments both synthetic real show superior performance of over competing methods offer insights trade-offs between factors.