Information Technology Reference
In-Depth Information
Grid cell
Edge-grid cell
Depth image
Edge extraction
No-edge-grid cell
Fig. 6 Edge-grid cell and no-edge grid cell
make regular mesh layers in each frame. Figure 7(a) shows the rendering result of
the wire frame mode and the rendering result for a 3D surface. However, it is not
enough for the surface of a 3D scene, because there are serious distortions in the
region of edges as shown in the rendering result.
A boundary layer includes depth pixels in the edge-grid cell. In order to regu-
larly extract depth pixels, we employed four quad-tree modes and a full modeling
mode. After uniformly dividing each edge-grid cell into four sub-grid cells, we use
the full modeling mode when more than two sub-grid cells include edges. Other-
wise, one of the quad-tree modes is selected according to the location of the sub-
grid cell that includes edges. Table 2 shows the full modeling mode and four
quad-tree modes in the boundary layer: up-left, up-right, down-left, and down-
right quad-tree modes. Here, we extracted 10 depth pixels in the quad-tree mode
and 21 depth pixels in the full modeling mode. It should be noted that we can han-
dle serious distortions, holes, close to the region of edges due to the difference of
depth values in the boundary layer. For preventing holes, additional processing is
required to fill out them, as shown in Table 2.
A boundary layer is used to refine a 3D surface generated by a regular mesh
layer for the region of edges. Since most of the serious distortions are mainly oc-
curred in their areas, we should deal with the region of edges carefully. Figure
7(b) shows the rendering result of the wire frame mode for a 3D surface with both
layers and its rendering result for a 3D surface.
A feature point layer includes depth pixels in the no-edge-grid cell. While we
deal with the region of edges to generate a boundary layer in each frame, feature
point layers is to handle the region of no edges. Feature point layers are used to
enhance the visual quality of the region of no edges in the 3D scene. In order to
determine the influential depth pixels in the no-edge-grid cells, scores of all pixels
in the no-edge-grid cell are estimated using a maximum distance algorithm. The
most influential depth pixels are then gathered into the 1 st feature point layer.
Likewise, the second influential depth pixels are also gathered into the 2 nd feature
point layer; this process is repeated for all subsequent points. Figure 7(c) shows
the wire frame for the final surface generated by regular mesh, boundary, and fea-
ture point layers. We can notice the visual quality of the 3D surfaces is enhanced
by adding the depth information in feature point layers.
Search WWH ::




Custom Search