Abstract: An automatic video object plane extraction algorithm is proposed.At first,a binary moving mask is obtained by Gaussianity test based on temporal motion information,followed by a further test by establishing MRF model on double-scale neighborhood to achieve smooth and complete binary moving mask. Then,an improved watershed algorithm based on nonlinear transform is employed to segment moving regions.Finally,moving objects are extracted by performing ratio operation on spatial and temporal results. Experimental results validate the proposed algorithm.