The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
648 
If the CSF under all condition such as static and motion is 
defined, then the details viewer perceived can be predicted. 
3.2 Clustering Perception Information 
Because the rendered image Irender contains all the details that 
can be display on the screen, while the vision simulated image 
Ifiltered only contains the perceptible details. To detect the 
unperceived details, the Idifference is defined as: 
I difference ^render ^filtered 
The /difference takes the advantage of the dissimilar pixels in the 
I ren der and 1'filtered to indicate the unperceived details in the image. 
Because the gradation difference of the pixel is related to the 
changes of the spatial frequency and contrast sensitivity. As a 
result, the degree of the gradation difference is also 
corresponding to certain spatial frequency and contrast 
sensitivity range. In another word, the gradation difference 
reflects the viewers’ perceive capacity to certain details. 
Use the MLC (maximum likelihood classification) method to 
classify the pixels in Idifference into four classes: none change, 
slight change, moderate change and dramatic change. The 
relation between the visual perceive capacity and classes is 
show in table 1. 
Idifference 
HVS filtering 
CSF Feature 
Vision 
no change 
No filtering 
Lowest CS 
and SF 
Highest 
sensitive 
slight change 
Slight 
filtering 
Low CS and 
SF 
High 
sensitive 
moderate 
change 
Moderate 
filtering 
High CS and 
SF 
Low 
sensitive 
dramatic 
change 
Heavy 
filtering 
Highest CS 
and SF 
Lowest 
sensitive 
Table 1. The relationship between perceptual degree and the 
difference image 
According to the table 1, the dramatic change of the gradation is 
corresponding to the lowest sensitive details in the image while 
the no change to the highest sensitive details, therefore the 
result is the foundation of the following grating of the triangle 
perceptual importance. 
4. PERCEPTION-DRIVEN SIMPLIFICATION 
ALGORITHM 
4.1 Transmission of Perception Information 
After the classification, the imperceptible details should be 
transmitted to the model’s geometry, and the mapping relations 
between the 2D space and 3D space should be established. 
In the real world, vision is formed when the light reflects into 
the retina. Likewise, the 3D models are displayed on the screen 
through pipelines of the display hardware, which transform the 
vertexes to various coordinates. Finally, the reader images are 
generated and displayed. 
Figure 3. The principle of ray casting 
According to the principle of ray projection and the render 
pipeline, given the camera position and the projecting plane, a 
radial through the pixel can link to the primitive of the 
geometry. This method is termed ray casting (Figure 3). 
The projection relationship is established between image space 
and model space through ray casting. Perceptual information is 
passed to the primitives can causes that primitive is of 
perceptual information according with it in the image. 
4.2 Aggregation Operation 
Based on plenty of the experiments, Kobbelt’s group discovered 
that the quality of the simplification has little to do with the 
way choosing the new vertex, but rather the order of the fold or 
contracting. This order is defined under the rule of error 
measurement (Kobbelt et al., 1998). Therefore, they suggest 
paying more attention to find the proper measurement. 
Although the minimized geometric error guarantees the fidelity 
of the simplified model, and the results are also predictable. The 
perceptual differences caused by the operations of 
simplification are ignored, which means the perceptual 
differences can not be predicted in the simplification. Be similar 
with the assumptions as the geometric error measurement, we 
make the assumption that if the perceptual error is the minimum 
during every simplification step, then the total perceptual error 
minimizing is going to be achieved. As a result, we propose that 
the simplify operations should fully consider about the 
perceptual change of the 3D model, and integrate the geometry 
with the perceptual error to measure the budget of the 
simplification. 
Based on the QEM proposed by Garland, the geometric error is 
calculated using a Q-metrix, this paper adopts the ratio of the 
new vertex’s geometric error and the diagonal line length of the 
minimum bounding box (MBB) of the triangles as the 
geometric budget of the merging operation (v,,v 2 ) —> v . 
Cost G = 
v Qy 
L 
v (Q, +Q 2 ) v 
L 
(3) 
Where Costc= the simplified budget of the folding 
operation (v,,v 2 ) —» v ; 
L = the length of the diagonal line of the MBB