Joint depth map interpolation and segmentation with planar surface model

Abstract

Depth map interpolation and segmentation has been a long-standing problem in computer vision. However, many people treat them as two independent problems. Indeed, the two problems are complementary. The results of one problem can aid in improving the results of the other in powerful ways. Assuming that the depth map consists of planar surfaces, we propose a unified variational formula for joint depth map interpolation and segmentation. Specifically, our model uses a multi-label representation of the depth map, where each label corresponds to a parametric representation of the planar surface on a segment. Using alternating direction method, we are able to find the minimal solution. Experiments show our algorithm outperforms other methods.