Abstract: Since street images usually contain privacy contents like pedestrians,a global optimized spatial-temporal image completion method is presented for pedestrian removal.Firstly,a structure from motion(SFM) algorithm is applied to build the relationship between the target and the reference images,which guarantees the effectiveness of the proposed method in complex situations.Then,a Markov random field(MRF) model is built and the global optimization method is used for the target region completion.In this stage,the spatial and temporal completions are integrated into one process.And a genuine and visually consistent result is obtained through an automatically selected completion method.Experimental results on a wide variety of images are presented,which demonstrate the effectiveness of the proposed image completion algorithm for complex scenes.