In this paper,we propose a novel neighbor combination framework for the reconstruction of the atmospheric turbulence degenerated image sequence.To utilize the spatial and temporal redundancy,a neighbor vector sampling strategy in spatial and temporal domain is conducted relying on the modeling of the registered sequence.Then,a combinator of neighbor vectors is developed based on a resampling maximum likelihood model and a relative approximation.Relying on the neighbor combination and spatialinvariant deconvolution,a clear image is reconstructed.Experiments on real data sets demonstrate the effectiveness of this framework.