International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
matching research the similarity measurement is oflen a 
difficult and puzzled problem. How to define the measurement 
parameters and procedures is the most important step. At 
present there are all kinds of measurement parameters for 
feature comparison, but they are not universal and very 
sensitive to the shape of features. Aiming at this, the authors 
propose a new algorithm. The basic principle of new algorithm 
is making a buffer for a feature and computing the similarity 
through the length of another feature inside the buffer (Sui, 
2002). This is shown as Fig.l. Suppose we make a buffer for 
old feature F1 with a given buffer distance BufferDis. and we 
use the new feature F2 to compare the difference of two 
features. Obviously when the ratio of the feature F2 fall into the 
buffer created by the feature F1 is smaller, the feature F1 is apt 
to change. Vice versa. So the detection formula for line feature 
can be defined as the following: 
/ 
> .. outer 
Line — (1) 
Ltd 
In the above formula, p represents change ratio for line 
Ane 
feature. L,,,,,' represents the length outside the buffer, Lz... 
represents the whole length of feature F2. 
Similarly. for polygon features the formula is defined as the 
following: 
P Im Loser (2) 
Sirface — 4 
Total 
In the above formula, Psurface Tepresents change ratio for 
surface feature, A, represents the area outside the 
buffer, 4,.,,,.. represents the whole area of feature F2. 
Totalrr 
It can be seen that the changed degree can be controlled by 
adjusting parameter Pj;,,, or Ps,4;,,,. Generally they can be 
taken as 0.85. Obviously, this algorithm is not sensitive to 
shape of features and it is universal for all the features 
comparison. 
old feature buffer 
new feature 
    
New feature outside new Feature inside the 
the buffer buffer 
Fig 2 The principle of buffer detection algorithm. (The black 
middle line is old line feature and the blue dotted 
line is new line feature.) 
2.1.2 The formula for buffer detection distance 
One key problem for buffer detection is to how to compute the 
buffer distance. If ignoring the tiny errors (like data conversion, 
computing etc.), for change detection between new image and 
old map. the buffer distance is mainly dependent on the 
accuracy of origin old map and the registration between new 
image and old map. Suppose the RMSE of the detecting feature 
the RMSE of registration between old map 
in old map is 6, . 
and new image i$ O,,,,5,,,;,,. then the formula for buffer 
distance can be deduced as the following: 
Duel) mme van ale 1 o Eye A (3) 
registration map 
Similarly. for the buffer distance between new map and old map 
with same map scale, the distance is mainly dependent on the 
accuracy of origin old map and new map. Suppose the RMSE 
the RMSE of 
of the detecting feature in old map is Gold -map 
the detecting feature in new map is 6,4, 4, . then the formula 
for buffer distance can be deduced as the following: 
2 2 
BufferDis (4) 
map nap — O olg map * O new—map 
2.2 Double-buffer detection algorithm 
2.2.1 — The principle of double-buffer detection algorithm 
For feature level change detection (FLCD) based on old map 
with small scale and new map with large scale, it is necessary to 
consider the effects of cartographic generalization. As everyone 
knows, the shape simplification and generalization are 
implemented by many generalization factors like merging, 
splitting, exaggerate and so on. The basic principles for shape 
generalization are as the following (Wang, 1992): 
eo Keeping the shape similarity of main features 
e Keeping the accuracy of key feature points 
e Keeping the contrast between different curve 
segments 
This means that there exists quantitative relationship between 
two features before and after generalization. However, the 
condition satisfied with this kind of quantitative relationship 
should be that the difference between two map scales is small. 
Because of generalization the buffer detection algorithm cannot 
be suitable for detecting changes between new and old maps. 
However based on this kind of quantitative relationship, we can 
define two buffers: one buffer is created by old feature and 
employed for detecting the change of the new feature 
comparing to the old feature; and another buffer is created by 
new feature and employed for detecting the change of the old 
feature comparing to the new feature. The first buffer can be 
called front-buffer, shown as Fig.2(b). The second buffer can be 
called back-buffer, shown as Fig.2(c). The function of back- 
buffer is to detect the changes caused by generalization and the 
function of front-buffer is to detect the real objects changes. It 
can be seen that the whole changes can be detected completely 
through these two buffers. And based on two buffers this 
algorithm can be called the double-buffer algorithm. The 
detecting algorithm for the back-buffer and the front-buffer is 
same with the buffer detection algorithm. So the detection 
formula for line features can be defined as the following: 
outer -after 
Line- after = (5) 
Total -after : 
Lehre ; 
==> (6) 
Line-hefore I 
"Total - before 
D = x J4/ + 7 = 
/ Line — P enfe: W after + P, ine -before Ws (7) 
In formula (5).(6).(7) . P is change ratio of whole line 
Line S 
feature, P is the change ratio of line feature in back- 
Line-afier 
buffer, P 
; — is the change ratio of line feature in front- 
Line-before E 
460 
  
Interna 
buffer, 
factor 
feature 
of old I 
outside 
line fea 
buffer « 
(a) O 
2.2.2 T 
Differei 
detectic 
distance 
front-bi 
accurac 
for the 
factor f 
the bac 
front-bi 
RMSE 
32 CI 
map sc 
Change 
cannot 
main ef 
to quan 
features 
problen 
(Sui,20/ 
introduc 
all kind 
the oth: 
built ba 
general