Abstract: Multi-baseline phase unwrapping problem can be solved according to find the optimal solution of the L1-norm optimization. However, there are two problems: one is the huge memory required and the other is the difficulty in processing interferograms with severe noise. In order to decrease the memory requirement of the L1-norm method, with a cost function of L-norm is employed to approximate the L1-norm. Consequently, the objective function of the improved multi-baseline phase unwrapping is the form of L-norm+L1-norm, and the size of the new optimization variable is decreased by 57%. The performance of the proposed algorithm is validated via a real dataset with severe noise present, and the experiment demonstrates that the proposed algorithm not only presents a well phase unwrapping result of interferograms with good quality, but also performs a filtering against noise region.