In order to solve such problems existing in the traditional information divergence algorithm as the inconsistence of object image sizes and the low accuracy of the depth recovery in non-contiguous areas,a new information divergence algorithm is proposed based on space-variant window. In this algorithm,first,the homography matrix between two original defocused images is calculated. Then,by using this homography matrix,the original images are rectified to obtain new defocus images of the same size. Moreover,the depth of non-contiguous area is estimated through the space-variant window structure. The simulated and real-image experimental results show that the proposed algorithm can avoid the smoothing depth in discontinuities and improve the precision.