Section 3. And relevant image processing technologies are 
introduced in Section 4. Experiments are carried out on the 
given images and conclusions drawn from the results are 
presented in Section 5. 
2. IMAGE CHARACTERISTIC AND TARGET 
FEATURE ANALYSIS 
Figure 1 (a) and (b) are a SAR image and an optical image 
which have been registered, respectively. The imaging region is 
an airport, so the typical line-type target to be detected is the 
runway here. One can obviously see that the optical image fit 
the human vision well. The image is smooth as a whole. The 
runway is comparatively bright in the optical image because of 
its material. However, some route signs (seen in the top right 
comer of the image) on the runway and some trails (seen in the 
middle of the right runway) which are caused by abrasion or 
erode or some other reasons will present evident edges within 
the runway area and baffle the location and detection of the 
runway edge pixels. While, in the SAR image, those signs and 
trails “disappear”, because objects made of the same material 
has almost the same backscattering coefficient G for 
microwave. However, the speckle noise is evident in the whole 
image. It can be seen in the area between the two runways. (It 
also exists within the runway, but it is too dark to be seen there.) 
This phenomenon is a big challenge to SAR image processing. 
3. PROCESSING FLOW OF IMAGE FUSION 
Basing on the above analysis, a detection method for line-type 
targets using the SAR and the optical image fusion is proposed 
in this paper. The processing flow of this method in shown in 
Figure 2, and the denotation * indicates that the image is a 
binary image. The detailed procedures are as follows. 
(1) apply Canny detector (Canny [1986]) to the optical image in 
order to extract as much edge information as possible; 
meanwhile, adopt the segmentation algorithm (Robert A., 1998) 
based on Markov Random Field (MRF) to obtain the regions 
with weak backscattering coefficient in the SAR image; Canny 
edge detector is one type of edge detector though the linear 
smoothing filter which has strong resistance of Gaussian noise 
exiting in most optical images. MRF-based segmentation 
algorithm introduces the image intensity as well as the structure 
information into the processing, thus it can obtain appealing 
results when applied to SAR image with the speckle noise 
which is supposed to be multiplicative existing. 
segmentation 
based on MRF 
geometric 
characteristic 
morphology 
filtering 
Roberts edge 
deteour 
Figure 2 Processing Flowchart 
(2) further extract ROI out of the segmentation result according 
to the geometric characteristic of runway areas; 
Typical man-made targets, such as highways and airport 
runways, have distinct geometric characteristic in the image 
which will favour the detection of this type of targets. They are 
always very straight lines and their comers are usually obtuse 
angles or right angles. Since Hough transform (Hough [1962]) 
is commonly used in straight line detection, it is applied to the 
edge image of the segmentation result which is obtained by 
Roberts detector. The edge image is binary in xy-plane (row 
and column). Here a brief introduction of Hough transform is 
given. More detailed information can be found in some relevant 
references or books. The essence of the Hough transform is 
that p = x cos 0 + y sin 0 is used as a normal representation of a 
line. Its computational attractiveness arises from sub-dividing 
the pO parameter space into so-called accumulator cells. At the 
end of this transform, the image is transferred to pO -plane, and 
the value Q in the position (p. , 6 j ) means that Q points in the 
xy-plane lie on the line p. = x cos + y sin 6.. Therefore, the 
long and straight lines represent themselves by the strong points 
with big value in p6 -plane. Detect the peak points in pQ - 
plane can find out which edge is characterized by long and 
straight lines. Then, segmented regions with these detected 
edges will retained and the others will be discarded. And the 
ROI image is obtained. 
(3) filter the ROI image (which is binary) with morphological 
operator (opening and closing) in order to fill the holes that are 
caused by the airplanes whose backscattering coefficient is 
fairly strong within the runway area; 
Opening operation can smooth bigger objects and discard 
smaller objects. While closing operation is able to fill tiny holes 
internal to the object and connect adjacent objects. 
(4) apply the simpler Roberts detector to obtain the edge of the 
modified ROI image; 
Roberts detector is one of the oldest edge detectors in digital 
image processing, and it is also the simplest. Since a binary 
image is to be processed here, this detector is the most 
economical choice in terms of processing speed and simplicity. 
(5) eliminate irrelevant pixels in the optical edge image with the 
ROI edge of SAR image as the reference in order to get the 
final detection result. The fusion rule here is that search the 
neighbourhood in the ROI boundary image whose centre is the 
edge pixel in the optical edge image; if there is the boundary 
pixel in the neighbourhood, the centred edge pixel will be