Joint Depth Map Upsampling via Prior Minimum Spanning Tree

Abstract

Depth map is the fundamental of various 3D applications. While consumer-level depth cameras have gained much attention recently due to their affordable cost, the low resolution and bad quality of depth maps generated by them have obstructed their applications in practice. In this paper, we propose a novel depth map upsampling algorithm based on prior minimum spanning tree (pMST). Utilizing the registered color and depth images, we construct a prior map which indicates the probability of the co-occurrence between depth discontinuities and color image edges. Under the joint guidance of prior map and color image, a prior minimum spanning tree is extracted and the high resolution depth map is obtained via a joint bilateral filter on the pMST. Experimental results demonstrate the effectiveness of our method compared to existing local depth super resolution methods. Keywords—Depth upsampling; Minimum Spanning Tree; Edgeaware filter; Depth super resolution; Prior map