The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
647 
In summary, the perception-driven simplification method for 
complex buildings should extract the perception details from 
rendering images, not from geometrical characteristic. The 
geometrical characteristic of the model can’t offer credible and 
perception information, while the rendered images can not only 
reflect on features of the Human Visual System (HVS), but also 
the perceptual details of the model exactly. 
2.2 The Framework of Perception-driven Simplification 
As shown in Figure 1, there are 5 main steps in this method 
from the original complex building model M 0 to simplified 
model Mi : 
Figure 1. Sketch map of perception-driven simplification 
2.2.1 Acquire the Rendered Image 
It is possible to get the rendered image I ren der in certain view 
point by setting the complex building model M 0 and the light, 
camera parameters, place, the relational rendering algorithm in 
its virtual scene. It can all-sided detect the 3D building model’s 
perception information by using the rendered images under 
multi-viewpoints. 
2.2.2 Simulate the Human Eyes’ Perceptual Images 
Discrete wavelet transformation (DWT) can ensure a good 
multi-resolution analysing function as well as makes the CSF 
function, used to describe human eye perceptual characteristic, 
possible to map with subbands of frequencies in different 
ranges. 
2.2.3 Extract and Classify the Perception Information 
The difference image 1difference between image I render and If,n e red 
contains some details unobservable to human eyes, which can 
be classified into four groups of change, as tiny, light, moderate 
and heavy. Thus the change of pixels’ grey levels effectively 
reflect out the visual difference. 
2.2.4 Transmit the Perception Information 
By ray casting, the perception information will be transmitted 
to the triangles on the model surface. And the rank of change is 
recorded into the attribute of the triangle. 
2.2.5 Simplify Models 
According to the perception attribute of triangles, edge-collapse 
operation will be executed, so that the imperceptible details on 
model surface will be removed dramatically. 
3. PERCEPTION INFORMATION AQUIREMENT 
BAESED HVS FILTERING 
3.1 HVS Filtering 
The perception-driven simplification methodology of 3D 
complex building models applies the HVS to filter the image. 
Firstly, it simulates the image information human eye received. 
Then the perceptual knowledge is extracted by comparing the 
original render image with the vision simulated image. Finally, 
the perceptual knowledge can be used to conduct the 
simplification. The most important thing of above is the vision 
simulation. 
One of the most important issues in HVS application is to 
reduce the contrast sensitivity of the high spatial frequency in 
the image (Nadenau et al., 2003).The common used way is to 
compute and quantify the spatial frequency by Contrast 
Sensitivity Function (CSF) (Figure 2). 
0.001 
1 10 
spacial frequency (c/dcg) 
100 
Figure 2. Contrast Sensitivity Function (Reddy, 2001) 
The CSF explains the relationship between the contrast 
sensitivity and spatial frequency, and describes the range of the 
gratings that can be perceived by HVS. The contrast threshold 
is also given at certain spatial frequency and contrast for the 
viewers. 
Mannos and Sakrison propose a general formula (1) to compute 
the contrast sensitivity (Mannos et al., 1974), which has been 
testified and quoted by many other researchers(Meng et al., 
2004; Qu et al., 2006; Watson et al., 2004).. 
A(a) = 2.6(0.0192 + 0.144aK (0144o) " (1) 
Where a =the spatial frequency, taking cycle per degree (cpd) 
as the unit; 
A( a ) = the Michelson contrast