A Real-Time Depth Map Refinement and Disparity Ranges Expansion System (DRDE) for Multiview Rendering

Abstract

Glasses-free stereoscopic three dimensional (3D) displays is becoming a competitive alternative to traditional 2D display presentations. Depth sensation, however, is still a major problem for most existing 3D displays. Multiview displays usually exhibit a fairly shallow disparity of field, and often fail to present sufficient disparity ranges for realistic natural scene. Besides, multiview rendering by depth image based rendering (DIBR) requires high-quality depth estimates. However, the exiting depth refinement methods are computational extensive, real time rendering for display is difficult to achieve. In this paper, we present a practical depth-map refinement and disparity ranges expansion system (DRDE) technique designed specifically for multiview rendering in 3D display. We use a reference image guiding method to quickly refine the initial depth map. After our proposed depth refinement, the outlines of the objects in the depth are enhanced while the areas with less texture are simultaneously smoothed, resulting in high quality multiview rendering images. For a greater disparity range, we modify the traditional parallactic transformation, which is generated by a parallel geometrical model, to a rotation one, during the virtual viewpoint image rendering process. We carried out extensive visualization experiments, and the results showed that our algorithm significantly improved the depth sensation in final 3D display. Our multiview rendering system with 1080 x 1920 image resolution for auto stereoscopic multiview display is able to achieve a rendering speed as fast as 62.5 frames per second (fps) for 9 viewpoints.

DOI: 10.1145/2949035.2949037

Extracted Key Phrases

8 Figures and Tables

Cite this paper

@inproceedings{Pei2016ARD, title={A Real-Time Depth Map Refinement and Disparity Ranges Expansion System (DRDE) for Multiview Rendering}, author={Renjing Pei and Zheng Geng and Zhaoxing Zhang}, booktitle={CGI}, year={2016} }