International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
also get that the algorithm is robust to partly distortion from
table 2. In table 2, the object images are the transform of the
template images, but are distorted 5% in horizon and vertical
respectively.
Template | Circle | Square Oblong Triangle
Object (1:2)
Circle 0.1052 | 3.4109 77127 9.9295
Square 3.4109 | 0.0539 3.7840 4.4781
Oblong (1:2) | 7.7127 | 3.7840 0.1038 6.6609
Triangle 9.9295 | 4.4781 6.6609 0.3828
Table 1. Correlation results of multi basic shape based on DPC
Circle | Square | Oblong(1:2) | Triangle
Circle 1.939
Square 1.9394
Oblong (1:2) 1.9109
Triangle 1.3482
Table 2. Correlation results of distorted images and templates
5.3 Vehicle extraction and recognition
Here we give an example to show vehicle extraction and
recognition from remote sensing image based on hierarchical
template, figure 7 is the wavelet and morphological transform
result of figure 6. From image in figure 7 We acquire the
interested region in the input image, we choose a small window
(figure 8) to do the further processing. first we extract feature of
the vehicles in the window image in two steps: 1) edge
detection using canny algorithm (figure 9), 2) edge tracking to
find the profile of the vehicles by combining the gradient and
gradient direction (Hang, 2000; Shaoqing, 2002). Secondly, we
correlate the profiles with vehicles profile template by DPC sets
and most the profiles extracted from the image has good
correlation with the vehicles oblong template although the
image profiles have some distortion. From this we can get a
whole concept where the vehicles most possibly are. In those
profiles that has good correlation with the templates we try to
find the further features such as windshield, and the size of the
profiles according to the resolution and number of the pixels
contained in the profile. Then we can get a further recognition
of the vehicles in the image. The results show that we can
recognition almost all the vehicles in the input image.
Figure 6. Input example image that contain highways, vehicles,
airplanes and so on.
Figure 8. Small parts of Figure 9 Canny ' edge
input image detection results of left image
6. CONCLUSIONS
Objection recognition in remote sensing image is a challenging
work. In this paper, we try to establish a hierarchical template
for the object to recognize the objects from common features to
particular features. it is time-saving method since it avoids
searching whole input image with complex features. A method
for template correlation based on DPC set is also presented in
this paper and the experiments show that it is effective for
correlation between the profile of the object template and
image.
7. REFERENCES
Ballard, D.H., 1981. Generalizing the Hough transform to
detect arbitrary shapes. Pattern Recognition, 13(2), pp. 111-122.
Brown, L.G., 1992. A survey of image registration techniques.
ACM Computing Surveys, 24(4), pp.325-376.
Hang, D., Yu, Y. Jun, S. Songyu, Y., 2000. Snake Model for
Edge Detection , Journal of Shang Hai Jiao Tong University,
Vol.34, No.6.
Rensheng, W., Xiaoguang, J., Jianlin, Z., 1997. A Studyon
Detecting Man-made Objects from Natural Background in
Space Sensing Imagery, Journa lof Image and Graphics. Vol.2
No.7.
Rucklidge, W.j., 1997. Efficiently locating objects using the
Hausdorff distance, /nternational Journal of Computer Vision,
24(3), pp. 251-270.
Selvarajan, S. and Tat, C.W., 2001. Extraction of man-made
features from remote sensing imageries by data fusion
techniques. In: The 22" Asian conference on remote sensing,
Singapore.
Shaoqing, Y. and Chuanying, J., 2002. Image edge connection
based on fuzzy logic, Optical Technique, Vol.28, No.2.
KEY
ABS
The ;
ing f
cont
sevel
The :
An |
phot
thors
was.
They
sor |
Prof
Verc
Gian
lerm
Radi
The
inert
quire
racy:
tion;
and/
intrii
asses
lead
aide
A qu
the 1
sion
more
GPS
last
carry
draw
form
1:50
the
with
The
amir
orie;
Betv
dam:
