74
Michela Bertolotto
Indeed, an interesting property of some techniques for progressive transmission
of raster data (e.g. interleaving techniques) is the fact that they only transmit increments (i.e. sets of new pixels) at each step. Such increments are added to the currently displayed image to improve its resolution without requiring the transmission
or downloading of another complete image version.
Since vector files can be very large, it would be useful to speed up the transmission in a similar way and thus reduce network traffic. We have tried to incorporate
this property into our proposed data structure. Instead of storing and transmitting the
entire intermediate map representations, only the coarser map version is fully stored
while subsequent levels just store increments (i.e. in this case, newly introduced entities and refined representations of entities present at previous LOD). An example
is illustrated in Fig. 4.4. An important aspect of our data structure is that vertical
links between different representations of the same polyline are explicitly encoded
and transmitted (see [51] for more details).
Note that as intermediate levels are not fully stored in the database, they have
to be reconstructed on the client site by merging increments into the coarser fully
stored map version. However, increments between two consecutive levels of detail
of a vector data set are not simply sets of pixels. They can be complex sets of entities
that are added to the level at finer detail to refine the representation of a set of entities
at the lower level. The integration of such increments with the currently downloaded
or displayed representation might be a non-trivial task if topological consistency
between different representations must be preserved.
In the case of sequences of maps generated by a topologically consistent line
simplification algorithm (such as Saalfeld’s algorithm), the reconstruction process
is relatively simple as topological relations between different entities do not change
(the endpoints of polylines remain unchanged, only their shape varies).
We have performed experiments on the progressive transmission of these multiple map representations using real data sets [52]. Results showed that transmitting incrementally the entire sequence takes approximately (with a difference of
very few seconds on average) the same amount of time as sending the fully detailed map in one step. An example is reported in Table 4.1, where a hierarchy with
three different levels is considered. The specific data set that produced these results
contains over 30,000 lines represented by over 70,000 points (see [52] for more
details).
The experiments were performed using a server with 1.6 GHz CPU processor
and 2 GB memory and a client with 2.4 GHz CPU processor and 512 MB memory.
We measured the average response time for transmission via LANs at the speed of
100 Mbps.
Table 4.1 shows that, when the map is transmitted progressively, users can start
working with a coarser representation downloaded from the server within half of the
time required to download the full resolution map; such a representation is then gradually refined until an acceptable version is obtained. Even if the user is interested in
the fully detailed map, the response time is not increased significantly (one additional
second in this case). However, we expect that frequently the time of transmission
would be even further reduced, as in general users would not need to download all
Michela Bertolotto
Indeed, an interesting property of some techniques for progressive transmission
of raster data (e.g. interleaving techniques) is the fact that they only transmit increments (i.e. sets of new pixels) at each step. Such increments are added to the currently displayed image to improve its resolution without requiring the transmission
or downloading of another complete image version.
Since vector files can be very large, it would be useful to speed up the transmission in a similar way and thus reduce network traffic. We have tried to incorporate
this property into our proposed data structure. Instead of storing and transmitting the
entire intermediate map representations, only the coarser map version is fully stored
while subsequent levels just store increments (i.e. in this case, newly introduced entities and refined representations of entities present at previous LOD). An example
is illustrated in Fig. 4.4. An important aspect of our data structure is that vertical
links between different representations of the same polyline are explicitly encoded
and transmitted (see [51] for more details).
Note that as intermediate levels are not fully stored in the database, they have
to be reconstructed on the client site by merging increments into the coarser fully
stored map version. However, increments between two consecutive levels of detail
of a vector data set are not simply sets of pixels. They can be complex sets of entities
that are added to the level at finer detail to refine the representation of a set of entities
at the lower level. The integration of such increments with the currently downloaded
or displayed representation might be a non-trivial task if topological consistency
between different representations must be preserved.
In the case of sequences of maps generated by a topologically consistent line
simplification algorithm (such as Saalfeld’s algorithm), the reconstruction process
is relatively simple as topological relations between different entities do not change
(the endpoints of polylines remain unchanged, only their shape varies).
We have performed experiments on the progressive transmission of these multiple map representations using real data sets [52]. Results showed that transmitting incrementally the entire sequence takes approximately (with a difference of
very few seconds on average) the same amount of time as sending the fully detailed map in one step. An example is reported in Table 4.1, where a hierarchy with
three different levels is considered. The specific data set that produced these results
contains over 30,000 lines represented by over 70,000 points (see [52] for more
details).
The experiments were performed using a server with 1.6 GHz CPU processor
and 2 GB memory and a client with 2.4 GHz CPU processor and 512 MB memory.
We measured the average response time for transmission via LANs at the speed of
100 Mbps.
Table 4.1 shows that, when the map is transmitted progressively, users can start
working with a coarser representation downloaded from the server within half of the
time required to download the full resolution map; such a representation is then gradually refined until an acceptable version is obtained. Even if the user is interested in
the fully detailed map, the response time is not increased significantly (one additional
second in this case). However, we expect that frequently the time of transmission
would be even further reduced, as in general users would not need to download all