(a) image (b)Histogram thresholding 
segmentation 
(c)K-means segmentation (d)Multi-scale segmentation 
on SOM 
Figure 4.Point cloud image and segmentation 
Fig.3(b) is a segmentation result of Fig.3(a) by histogram 
threshold value algorithm. In Fig.3(b), the contour of road 
network are clear, but the water area in the lower right comer 
can not be partitioned. In Fig.3(c), the segmentation result show 
that K-NN algorithm has some smooth effort of image noises 
and the contour of road are also clear. By analyzing, we find 
that using histogram threshold value algorithm and K-NN 
algorithm is great effective for extracting geographic objects 
with greater contrast of radiant intensity. Because of the greater 
contrast between the road and the background in Fig.3, the 
segmentation contour of road network are clear in Fig.3(b) and 
Fig.3(c). In Fig.3(d), the image is segmented by using SOM 
based on the directional wavelet transform at multiple-scale. By 
comparing the different result, we find that SOM segmentation 
prior mark winner of the competition, and the lateral inhibition 
keep the nonlinear connection at the same time. According to 
the effect of the segmentation contour of road, it is found that 
the class flag is weakened gradually in the topologic 
neighbourhood of victorious features. That is to say there still 
are some activated features around the contour of road class. In 
Fig.3(d), the water area in the lower right comer is segmented 
equality ,clearly and accurate. It is indicated that the SOM 
segmentation is obviously effective for extracting the block 
objects. 
Fig.4(b) is a segmentation result of Fig.4(a) by histogram 
threshold value algorithm. It show that histogram threshold 
value algorithm has good effect on segmenting images which 
have greater contrast of intensity between objects and 
background. Fig.4(c) is a segmentation result by K-NN 
algorithm. It is observed that K-NN algorithm has some smooth 
effort of image noises and the contour of ancient architectural 
structures are clear. In Fig.4(d), the image is segmented by 
using SOM based on the directional wavelet transform at 
multiple-scale. According to foregoing analyzing, there is a 
certain influence among the features of SOM segmentation and 
activated features still exist in the topologic neighborhood of 
victorious features. SOM segmentation is good for extracting 
the block objects 
5. CONCLUSION 
There exist some problems in multi-scale wavelet transforms 
based on Mallat fast transform of discrete wavelets, in which 
image features may be only acquired in horizontal, vertical, and 
diagonal directions, and in k-nearest neighbor (k-NN ) 
segmentation the relation between image features are not 
considered. We propose a novel approach in which remote 
images may be segmented in two steps based on the directional 
wavelet transform at multiscale. Firstly, the images are 
transformed by the directional wavelet at coarser scale, multi 
dimension feature vectors are constructed, and the images are 
coarsely segmented by k-NN algorithm. Secondly, the result of 
coarse segmentation is used for prior messages , at fine scale 
the image segmentation is performed again by self-organizing 
map algorithm(SOM). In this paper, airspace and 3D-laser 
cloud points image are segmented by histogram threshold value 
algorithm, K-NN algorithm and our method. The result of 
experiment show that our approach is obviously good for 
extracting the block object. The disadvantages of SOM 
segmentation is that the computation cost much time. The 
reason is that it orderly update victorious features in every time 
of iteration(serial algorithm), but K-NN update all in every time 
of iteration(parallel algorithm ). At present, the improved 
research about SOM parallel algorithm is being studied. 
Because of extracting the geographic objects in remote sensing 
images is a emphasis of the research about remote sensing 
technology and application. The emphases in the farther study 
are followed: algorithm and adaptability about different remote 
sensing images and geographic objects, the image construe and 
algorithm based on knowledge, algorithm about extracting 
geographic objects based on multi-sensors remote sensing 
image fusion, the image construe and algorithm about multi 
band remote sensing image. 
6. REFERENCES 
6.1 References and/or Selected Bibliography 
References from Journals: 
C.A. Bouman, M. Shapiro. A Multiscale Random Field Model 
for Bayesian Image Segmentation [J], IEEE Trans, on Image 
Processing, 1996,3(2): 162-177 
C.Fosgate, H.Krim, W.Iring, et al. Multiscale Image 
Segmentation and Anormally Enhancement of SAR Imagery[J]. 
IEEE Transactions on Image Processing, 1997,6(l):7-20 
C T Li. Multiresolution Image Segmentation Integrating Gibbs 
Sampler and Region Merging Algorithm[J], Signal Processing, 
2003,83:67-78 
F. Pemkopf. Bayesian network classifiers versus selective yfc-NN 
classifier.[ J], Pattern Recognition , 2005,38(1): 1-10.