The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B6b. Beijing 2008 
then we can restore the line under scale scale j. Fig 4 show the 
process of line rebuilding. 
5. EXPR1MENTS 
Compared with models in [3][4], the model proposed in this 
paper owns following characters: 
Table 1. We use two line data packages in this experiment. The 
data volumes of the two data packages are 1.9 MB and 3.2 MB, 
and the data volumes of the first level after simplification are 
0.57MB and 0.96MB respectively. We measure the response 
time of user request to the coarsest level of data for the first 
time. For the first data package, the response time of our model 
is 0.42s while that of MSLT is 1.03s. Since there is less 
database 10 cost, it need less response time in our model. 
1) The models in this paper support editing. We use related id 
instead of absolute id to build vertical index in the line to 
represent order of vertices in the line. When new vertex is add 
into the line, we just need to modify several nodes in the multi 
scale line model in local area, without affecting the whole 
structure. 
2) Efficient database access. For the models in [3][4], even we 
just request the line in the coarsest level, it still must get data 
with all scale level. Spatial data are usually stored in the fields 
of BLOB field. The efficiency to access the BLOB field is low. 
The data in different level in our model are independent of each 
other. Therefore, data in different level can be stored into 
different tables or tablespaces. When users just need data with 
coarsest scale, we just need read data of that level. 
In this paper, we developed a prototype system for 
generalization and representation of multi-scale line. The 
system simplified the lines into three levels. The results are 
showed in Fig 5. We can find the lines in the third level are 
very close to the shape of original lines while the data volume is 
just 31.4% of original one. So we still can get good visual 
effects when users firstly request data and get satisfying web 
response speed, which is meaningful to mobile terminals such 
as cell phones and PDAs based on the wireless network. 
From Fig 5 we can also find topological relationships are also 
well persevered. 
(c) Third Level 
(d) Original Line 
Figure 5 Multi-scale line of four levels 
Table 1 response time of retrieving and rendering of data of 
coarsest level between MSLT and our model 
From above table, we can draw such conclusion that our model 
is more efficient in retrieving and render coarsest data for the 
first user access. 
6. CONCLUSION 
Scale is anther important property of spatial data besides 
geometry and attribute. In this paper, based on the analysis of 
spatial characters of spatial lines, by Visvalingam-Whyatt 
algorithm, we simplify spatial line into different scales. In this 
paper, we present a multi-way tree based multi-scale line model 
to store and manage line information under different scales. By 
increment data, the simplified lines can integrate with 
increment data to restore original data. Compared with Strip 
tree, our model can clearly manage data under different scales. 
Compared with MSLT of Jones, our model supports edition of 
multi-scale line, and own high database access efficiency. 
REFERENCE 
1. Buttenfield, B.P. Transmitting Vector Geospatial Data across 
the Intemet[C], Proceedings GIScience, Lecture Notes in 
Computer Science, Vol. 2478. Springer-Verlag, Berlin. 
2002:51-64. 
2. Ballard D H. Strip Trees: A Hierarchical Representation for 
Curves. Communications of the ACM. 1981, 24(5):310-321. 
3. Chirstopher B. Jones, Ian M. Abraham. Design 
considerations for a scale-independent cartographic database [C]. 
In Proceedings 2nd International Symposium on Spatial Data 
Handlinig, Seattle 1986:348-398. 
4. Min Zhou, Michela Bertolotto. A Data Structure for Efficient 
Transmission of Generalised Vector Maps[C]. International 
Conference on Computational Science 2004: 948-955. 
5. Visvalingam M, Whyatt J D. Line generalization by repeated 
elimination of the smallest area. The Cartographic Journal, 
1993;30( 1 ): 46-51. 
We compared the response time or our model and that in [3] for 
retrieving the data with coarsest scale from database and render 
on the screen. The experiment environment is PHI 800, 512M 
RAM, SQL Server2000. The experiment result is showed in