Information Technology Reference
In-Depth Information
Figure 11 shows the results of 3D scene reconstruction on ROI of BEAR im-
ages using hierarchical decomposition. When we subjectively compare the scenes
with the one generated with the ground truth for BEAR images, the 3D scene gen-
erated by our method more closely resembled the original scene than other meth-
ods. Especially, the regions of the big bear doll's leg and the small bear doll in the
original scene were much similar with ours. Hence, we subjectively notice that
the depth image obtained by the hybrid camera has more reliable depth data than
the other methods. Figure 12 shows the result of depth estimation with the
ACTOR images and the side view of its 3D scene reconstruction. We could notice
that the regions of a table and a hat marked by circles were reconstructed into 3D
scenes better from depth information generated by the hybrid camera system than
the other methods.
(a) Belief propagation
(b) Gra ph cuts
(c) Dynamic programming
(d) Hybrid ca m era system
Fig. 12 Results of depth estimation for ACTOR image
5.2 3D Video Contents Generation
In this example, we have created 3D video contents with the main theme of a
home-shopping channel scenario using the hybrid camera system in the MPEG-4
Search WWH ::




Custom Search