Bi-Directional Depth Propagation for 2D-to-3D Conversion with Color/Depth-Based Superpixel Segmentation

摘要

In this paper, we propose bi-directional depth propagation for 2D-to-3D conversion with color/depth-based superpixel segmentation. Depth propagation generates the depth image of the query frame based on motion information between two sequential frames. However, color-based superpixel segmentation causes segmentation errors due to weak edge and analogous colors, thus resulting in motion estimation errors. We provide color/depth-based superpixel segmentation for accurate motion estimation instead of the color-based superpixel segmentation. Experimental results show that the proposed method successfully estimates motion vectors for depth propagation and outperforms state-of-the-arts with comparable time cost in terms of PSNR.