The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
1006 
(a) 
(b) 
(c) 
(d) 
Figure. 1 Several edge detection algorithm comparison (a) 
Removal noise Image (b) Robert operator processing image (c) 
Sobel operator processing image (d) Canny operator processing 
image 
2.3 Boundary Tracking 
From the edge of the target tracking features can be divided into 
two steps: Firstly, extract the basic unit-Edge reflecting the grey 
changes, secondly, connect these edges to meaningful goal, 
called edge tracking, that is, one searching process that is used 
to determine the contour images (Filtering refers to the image). 
Edge tracking is a method of border examining. Generally 
speaking, includes three processes: 
1. Determine the start point of search, which is strongly 
depended on this algorithm. 
2. Take an approximate data structure and search 
mechanism, determine the next border points on the basis 
of founding the starting point 
3. Termination of the search conditions. Carrying 
through edge tracking, first of all, find the edge of starting 
point. Making an edge point as a starting point, which have 
eight adjacent points. The central of edge point has the 
structure in Figure 2, can be as starting point of search. 
Search method: 
1. Choose a starting point of search and note the 
coordinates; 
2. Change it to 0, avoid duplication; 
3. Judge the present point of its eight neighbourhood in 
order, if there is 1, then set it as present point, output 
the direction this point, turn the step 2; 
4. Until there is no 1 in the eight neighbour regions to 
the current point, finish the track. 
DEE Ci 
Figure.2 The sketch of staring point for tracking border 
Edge tracking takes 8 neighbour templates, which is a method 
of record chain memorizing edge point, and regulate the 
required direction from current point to the next, because there 
are eight possible directions, which can be marked from 0 to 7 
and Figure 3 shows the coding schemes of the direction of an 
eight directions. Figure4 shows the border tracking results. 
3 
ft 
; 
k 
1 
n 
U 
\ 
/ 
>0 
/ 
K 
E 
5 
' 
7 
Figure. 3 Eight direction coding scheme 
2.4 Extraction of linear feature 
Before the interpolation, firstly, use the iterative algorithm of 
border demarcation to meet the fitting line of a straight line. 
Figure 5 is below. 
Figure.5 straight line split fitting Sketch 
While is splitting, the first, take the starting and ending points of 
connection as reference, in the range AB of curve, process the 
point-by-point search, get the P point of largest distance from 
AB, P’s distance equal to d, set threshold T, if d < T, the line AB 
can be used as fitting straight line. If d>T, P will be a split, and 
take AP and PB as new reference line, and repeat the process 
until there is no longer appearing the reference point of a 
straight line distance of more than T. 
In the sequence of broken line after splitting, a connecting credit 
line is composed of a straight line split in accordance with 
paragraph from the first to the end, beginning search from the 
first broken line, with the linear least squares fitting above 
mentioned, if the standard deviation between the next 
adjacent line less than a certain threshold value ofcp,then merge 
the two broken lines. Then combine the third broken line to 
judge, if the standard deviation between them less thancp,can be 
carry out the merge down, if more thancp,only merge the first 
and the second broken line, and then start search from the third 
broken line, process Ibid. Until the final sequence of broken 
lines, this can get the sequence of merge, each of which is from 
the fitting form of a straight line. In this paper , compared with 
other algorithm and on the basis of multiple experiments, 
choose T=3, 0=2 and get the desired results. For example 
Figure 6.