Image Processing Reference
In-Depth Information
has to receive and display the remote rendered image. This means only the display
stage is performed on the client, whereas all other stages are executed on the server.
Applications like SAGE [
23
] or VirtualGL
1
use this approach. Image streaming has
its main limitation in the presence of interaction delays and it is sensitive to network
latency. Nevertheless, this solution requires modest client resources and is suitable
for dynamic scenes.
In traditional 3D graphics systems, the scene consists of geometric primitives with
attributes like materials, colors and a set of lights. Based on these scene descriptions
the rendering system generates and displays an output image. Image-based rendering
(IBR) systems differ from traditional 3D graphics systems in that the client generates
images for the actual camera position based on pre-generated images rendered at dis-
cretized viewpoints. By using image-based rendering in a client/server architecture,
only the display stage of the visualization pipeline is executed on the client. The com-
plete rendering is performed on the server side with the traditional rendering. The
major advantage of image-based rendering is that the cost of viewing the scene inter-
actively on the client is independent of scene complexity and modest computational
resources are sufficient.
Apple's QuickTimeVR [
5
], an early IBR system, used a 360
◦
cylindrical envi-
ronment map to quickly generate an overview of the scene on the client. The main
limitation of using simple environment maps or panoramic images is that the view-
point is fixed in space. View interpolation or warping methods address this problem
[
31
,
32
]. Based on a depth value for each pixel, it is possible to warp the image to any
desired viewpoint on the client. The key challenge using interpolation or warping is
to fill the gaps of previously occluded areas which are now visible in the requested
view (see Fig.
31.10
). Another popular IBR approach is to use light fields [
26
]or
lumigraphs [
4
] where a discretized volume is formed by an array of images. The
client can now extract a slice out of this volume from any requested position and
orientation. Because many images are used for extracting a slice out of this volume,
the produced error can be reduced. Using this approach in a client/server architec-
ture requires the generation and transmission of multiple images for the client's slice
extraction.
Model-based rendering (MBR) systems in a distributed environment transfer raw
data, derived data or geometric primitives instead of images. This means that at least
the rendering and display stages of the visualization pipeline are performed on the
client side. While data or geometries are available locally, the client is given more
freedom inmanipulating viewing parameters and it may also reduce network traffic in
the end, but they require much more memory and rendering capacity often exceeding
the client's capabilities. Common post processing applications like Paraview, VisIt,
Ensight, FieldView or TecPlot use this approach and some of them are able to switch
to image streaming when local capacities are exceeded.
Hybrid rendering therefore try to optimize or balance computing and rendering
requirements between the client and the server. In most cases, parts of the geometry
are rendered by the client itself with low-resolution and high frame rates . The server
Search WWH ::
Custom Search