Hyperspectral and multispectral image fusion has been extensively studied in remote sensing. However, due to high-order spatial-spectral correlation, it is difficult to extract and approximate the fusion subspace across the modalities. In this paper, a new problem formulation is proposed to explore the intrinsic low rank correlation on spectral, and non-local self-similarity domains. Spectral-spatial structures with low rank constraints are investigated. The alternating direction method of multipliers-based method is adopted to solve the resulting fusion problem. Experiments conducted on the various datasets demonstrate the advancement of the proposed method in comparison with some the state-of-the-art fusion methods.