Abstract: To meet the requirements of speed and accuracy for interactive image object cutout,we propose to incorporate the generalized shape prior(GSP) knowledge,including the regional-GSP and the boundary-GSP,with the traditional color and gradient information by an energy function,and solve the energy function via graph cuts.The image object is extracted through a hierarchical procedure from global segmentation to local refinement.In the global stage,pixels are replaced by superpixels to preserve the image structures and dramatically reduces the computational complexities.The regional GSP is used to guarantee the accurate global results.The local process refines the errors in the boundaries of global result.It only considers the pixels in a subset around the boundary segment,which avoids the influence of other irrelevant pixels and makes the process more efficient.Furthermore,the boundary GSP helps to improve the quality of object boundaries.The results show that our method performs better in terms of speed and accuracy.