Depth hole filling using the depth distribution of neighboring regions of depth holes in the Kinect sensor

This paper proposes a depth hole filling method for RGBD images obtained from the Microsoft Kinect sensor. First, the proposed method labels depth holes based on 8-connectivity. For each labeled depth hole, the proposed method fills depth hole using the depth distribution of neighboring pixels of the depth hole. Then, we refine the hole filling result with cross-bilateral filtering. In experiments, by simply using the depth distribution of neighboring pixels, the proposed method improves the acquired depth map and reduces false filling caused by incorrect depth-color fusion.