In this paper,a conditional random field model based on spectral-spatial prior information (SSPCRF) is proposed to perform the task of change detection for high spatial resolution remote sensing images.The proposed method firstly introduces a saliency based sample selection strategy which considers the spectral-spatial information of observed difference image to improve the accuracy of modeling initial change detection result.Then a pairwise potential with boundary constraint is used to help keep the boundary of changed objects.Finally an inference method based on loopy belief propagation (LBP) algorithm is introduced to perform efficient optimization of the proposed model and get the final change map.The proposed SSPCRF model can greatly improve change detection accuracy while keeping detailed boundary information of changed objects.The proposed method is tested on two high resolution datasets and outperforms the commonly used change detection methods.