Autostereoscopic displays, allows one to see in 3D without glasses it requires at least 2 views and is often to maximize user comfort. However multi-view capture is too complex for more than two cameras. Furthermore there is no interoperability between 3D displays because each manufacturer has his specific input format. Thus we propose an efficient automatic conversion from stereoscopic views to N views for a rendering on various autostereoscopic screens. We describe the 2 main steps achieving automatic conversion from 2 to N views: disparity map estimation and view interpolation. We use graph cut algorithm to compute the disparity map with occlusions detection. As the quality of the disparity map is essential for interpolation we have developed an efficient management of occlusions through morphological operation, outliers erasing and padding with the smallest disparity. We thus reconstruct details even for concave objects. We also apply spatial regularization preserving borders and temporal smoothing with motion compensation. Those processes improve considerably the quality of the depths map videos. Virtual views are then generated through view morphing based on disparity to avoid 3D manipulation of data. Occlusion areas are managed and irregular morphing pixel locations are resampled with a B-spline filter adapted to view interpolation.

Date Published: 18 February 2009
PDF: 12 pages
Proc. SPIE 7237, Stereoscopic Displays and Applications XX, 72371H (18 February 2009); doi: 10.1117/12.806301

Published in SPIE Proceedings Vol. 7237:
Stereoscopic Displays and Applications XX
Andrew J. Woods; Nicolas S. Holliman; John O. Merritt, Editor(s)