The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
1007 
Figure.6 Linear Feature Extraction results 
2.5 Extract the main direction of buildings 
Extracted the main direction with main direction and its vertical 
line segment, excluded some segments that have nothing to do 
with the building edge, which can rule out some features of the 
surrounding buildings interference; extracted the main direction 
of buildings that is easy for searching the edge lines in this 
direction, improve the computing speed and enhance the 
stability of extraction. 
Assuming the total number of line segments is N, static the 
length of each line segment, as well as the corresponding 
azimuth. In the range of image dealing with, the building edge 
lines is connected much longer, so we should focus on the long 
segments and give much greater weight. In this paper, adopt the 
square of the length acting on the azimuth. Establish cumulative 
matrix, which will be cumulated to the corresponding angle, 
then the corresponding angle of maximum in the accumulator is 
the main direction of buildings. Figure 7 shows the effects of 
line planning. 
Figure.7 The effects of planning line 
Deletes line segments when Satisfy the following conditions: 
a € [0,90], |a-6?| > T and |a+90°-£?| > T (1) 
a € [90 180), [x-0|> T and \<x-90°-&\>T (2) 
Reserve line when Satisfy the following conditions: 
a € [0,90 ], |a-9\< T or \a +90 °-0\ < T (3) 
a € [90 180), \x-8\< T or| a -9O°-0|< T (4) 
2.6 Judge of segments relation with model. 
Because of noise or other buildings, the straight edge line of 
building may be cut off, there are many interferential segments. 
Which have an impact on the accuracy of building structures 
search. In this paper, with the method of model to judge the 
relationship of segments, connect the broken border line, and 
remove some interference surrounding the buildings. 
2.7 Handling the relation between lines. 
The long segment should be the main edge after connected the 
straight line, and there are some short line segments in the 
vicinity which can be considered as illusive edge line, they will 
be removed. First of all, sort all segments in accordance with 
the length, starting with the shortest straight line. Only proceed 
between the straight lines in the parallel. Figure 8 shows the 
effect of treatment. 
Figure 8 the effect of treating line relations 
2.8 Region segmentation and growth. 
Calculate the average gradient of each segment, sort the 
segments from the big to small according to the gradient, then 
extend the segments and segment the image to different region 
and mark each region. 
Region growth is a process that pixels or sub-regions are 
converged to bigger region in accordance with definitional 
criteria. If direct making use of region growth in the buildings 
within the seed growth, easy access the brink of noise, the edge 
of extraction is not linear. However, on the basis of extracting 
the edge of building, and processing segmentation, which 
relatively easy solve the noise problem and non-linear margin. 
2.9 Region merger. 
Calculating the average of grey U { and variance of each 
region, if the seed selection region meet — w o | < Au o and 
|<5^- — S Q | < AS q ,combine the growth of region. Au q and 
are u o S Q linear functions, the selection of parameters of 
function is due to the real case of buildings. 
After merging the top of buildings, extract the outer edge and 
track it, and now the result of tracing out at the edge is regular 
closed line, due to the restriction of straight line above, so that 
we have the rectangle between the adjacent edges, then act on 
straight line approximation, finally, we obtain the line on the 
top of buildings, which is the result of building extraction. 
Figure.9 The drawing of buildings edge extraction