Depth discontinuities are the layers of the world and is used to track object, image retrieval, evaluation of the 3D visual fatigue. To extract depth discontinuities from disparity map, we adopt DIP operator that is the difference of inverse probabilities for extracting sketch features that contain valleys and edges in 2D image. However, DIP needs a threshold to determine depth discontinuities and this is threshold needs user interaction. In this paper, we propose the method that extracts depth discontinuities without user interaction using learned threshold. We learned the threshold with the Middlebury disparity maps and made the evaluation to decide the threshold.