The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
649 
4.2.1 Perception Weight Factors 
In the simplification process, those operations on geometrical 
such primitives as vertices, edges and surfaces have not only 
brought in geometrical errors, what’s more, the results after 
geometrical transformation can cause perception diversity 
Therefore, this method chooses the sum of the change degrees 
between the new vertex brightness and those in the vertex-pair 
as the weight and makes adjustment only by geometrical cost. 
This article has chosen the complex timber-frame building as its 
testing object, most of whose appearance is made up of walls, 
cements, woods and so on which do not have highlight effect 
and this is rightly fit for the characteristic of Lambert 
illumination model. So, according to the analysis of light setting 
in the virtual scene, in the effect of three fixed spotlights, the 
expression to calculate vertex v brightness by using Lambert 
diffuse reflection model is: 
«=3 (4) 
i=l 
Where f, = attenuation factors 
I H = light intensity 
L t = incident direction’s unit vector 
In the process of confirming vertex-pair contracting (v,,v 2 ) —» v , 
it separately calculates vertex v, , v 2 and the new vertex v 
average normal along with its brightness, after that the 
perception weight of the geometric cost in the contracting 
process is: 
5. EXPERIMENTAL ANALYSIS AND DISCUSSION 
5.1 Description of Test Data 
A 3D model of Chinese classic building is chosen in the 
experiment, which has many components. This type of 3D 
model takes its characters, the model is made up of components 
having tenon. The Chinese ancient buildings’ structure is 
mainly formed by pillars, beams, arches and tiles, where every 
component is jointed with a tenon, composing bouncy frame. 
And this type of joining in timber-frame buildings has shaped 
plenty of depressed characters in the components, which 
benefits a lot to correctly meet and fix all the components. But, 
this tenon structure is entirely needless for real time 
virtualization and building its LOD model, thus it becomes the 
chief aim for simplification. 
5.2 Experimental Results and Discussions 
A prototype is developed to create multiple levels of building 
models. As analysed above, this article chooses tracing results 
from nonconservative strategies to guide the simplifying 
experiment for the whole model. 
During it, the whole triangles, which had a possibility to be 
simplified, acquired by the tracing results from all the effective 
elements of the rendered image account to 97,266, it has used 
the acuity degree shown in Figure 4 as the foundation to guide 
the simplification work, and then gained simplified result in 
Figure 4(c), where you can see the number been simplified 
accounts to 11,074 so the simplified rate is 11.39%. In Figure 4, 
it is clear that the model simplified driven by apperception has 
kept a high comparability with the original one. And the 
comparison in amplificatory details in Figure 4 (b)(d) has 
effectively explained the model has been simplified according 
to pre-assigned perception information. 
(5) 
Which separately weight the change degrees of the new 
vertex v brightness and vertex v, , v 2 brightness, where it may 
come that the brightness of new vertex turns to be lower than 
its original, I- < 7 V] or I- < / v2 , so it chose the absolute value 
of perceptive weight gene to eliminate the influence of the 
minus value calculated. 
After the analysis and calculation above, it separately defines 
the geometric cost and perception weight for the vertex-pair 
contracting (v 15 v 2 ) —» v , and then the cost of the vertex-pair 
contracting operation is: 
Cost 
(v.,v2 X v) =<P COSt G 
(6) 
The perception weight factor tp based on the change of vertex 
brightness can reorder the simplification operation by adjusting 
vertex-pair contracting cost, consequently realize optimization 
for the whole simplification operation of the triangle cluster. 
< a ) The animal model i b i détails ef the original model 
( c > simplified model * d ) details of simplified model 
Figure 4. The effect of simplified model 
Corresponding to perception ranks of the image elements, the 
triangles’ perception information can be classified into four 
groups, while only the heavy, moderate, light ranks need to be 
simplified, so that there are only three kinds of simplifying 
strategies, namely: First stair simplification, only to do with the 
triangles of the rank heavy, similarly, the second stair is for the 
rank heavy and moderate', and the third stair for all of three