Manifold-defect depth-map restoration for very low-cost S3D videos

Abstract

In this paper, the proposed algorithm provides a fluent and efficient method for repairing very-low quality depth maps of considerable manifold defects for low-cost stereoscopic 3D (S3D) photographing that such a depth map can be easily yielded by 1st-generation Kinect (Kinect-v1). The corresponding framework cascades two repairing portions named discriminative non-segmentation hole filling and edge rectification-by-deforming. The former can discriminatively fill a variety of depth-invalid holes with no need of practically making attribute-discrimination and target segmentation for depth holes. The main portions of the latter contain edge-shifting-rectification and texture-edge guided dual processing for tailoring possible twisted depth edges. Since the ingredients of proposed algorithm are compactly concatenated according to intimate context, most troublesome defects in Kinect-v1 depths can be tackled. Particularly, the proposed algorithm can obtain the restoring coherency for successive depth maps. A series of experimental results under various photographing scenarios can demonstrate that a single Kinect-v1 device can fast tell on the development of very low-cost S3D imaging tool by the proposed algorithm.