The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
Figure 5..Result using algorithm in (Comaniciu 
and Meer, 2002) with default 
parameter. 
5. CONCLUSION 
A segmentation algorithm based on spatial and spectral 
information fusion is proposed in this paper. The algorithm 
leads to an improvement in segmentation of land cover classes 
having similar spectral surface. 
ACKNOWLEDGEMENTS 
This work was supported by the 973 program and 863 program 
of People’s Republic of China under Grant 2006CB701303 and 
2006AA12Z132, the Foundation of the excellent State Key 
Laboratory under Grant 40523005, and the National Nature 
Science Foundation of China under Grant 40601055,the 
foundation of post-doctor of China under Grant 20060390825. 
Figure 6. The result of our method 
REFERENCES 
Wikipedia,2007 Lab color space conversion. 
http://en.wikipedia.org/wiki/Lab_color_space (Accessed 28 
March.2008) 
Chengjun, L. and Wechsler, FI.,2003. Independent component 
analysis of Gabor features for face recognition. Neural 
Networks, IEEE Transactions on, 14(4): 919-928. 
Comaniciu, D. and Meer, P.,2002. Mean shift: a robust 
approach toward feature space analysis. IEEE Transactions on 
Pattern Analysis and Machine Intelligence, 24(5): 17. 
Figure 6 is segmentation result using our method» Figure 5 is 
the result from EDSION softer ware implementing the 
algorithm in the paper (Comaniciu and Meer, 2002), which is 
a combination of fixed band Mean Shift filtering and region 
merging. To keep the equality of comparison, only the fixed 
Mean Shift filtering procedure is employed with default 
parameter spatial band 7 and rang band 6.5. Then filtering result 
is segmented into several classes by K-means method. From the 
Comparison of Figure 6 and Figure 5, we can find that the fixed 
bandwidth method confused the land cover types with similar 
spectral information. In figure6, the confusion of classes similar 
in spectral information is reduced in some degree. In fact, only 
by a simple religion merging algorithm to merging the region 
whose pixels is less a threshed, for example 20.a better result 
can be achieved, (as shown in Figure 7) 
Figure 7 the result with merging minimum region 
Comaniciu, D., Ramesh, V. and Meer, P.,2001. The Variable 
Bandwidth Mean Shift and Data-Driven Scale Selection, 
Proceeding of the IEEE international on Computer Vision, 
Canada. 
Dell'Acqua, F. et al.,2004. Exploiting spectral and spatial 
information in hyperspectral urban data with high resolution. 
Geoscience and Remote Sensing Letters, IEEE, 1(4): 322-326. 
Dong-chen, FI.,1990. Texture Unit, Texture Spectrum, And 
Texture Analysis. Geoscience and Remote Sensing, IEEE 
Transactions on, 28(4): 509-512. 
Georgescu, B., Shimshoni, I. and Meer, P.,2003. Mean Shift 
Based Clustering in High Dimensions:A Texture Classification 
Example, the Ninth IEEE International Conference on 
Computer Vision. 
Jung, C.R.,2007. Unsupervised multiscale segmentation of 
color images. Pattern Recognition Letters, 28(4): 523-533. 
LeiGuang, W., Wenbo, W., QinLing, D. and Qianqing, Q.,2006. 
Remote sensing image texture classification based on Gabor 
wavelet and support vector machine, SPIE. 
M.P.Wand and M.C.Jones,1995. Kernel Smoothing. Chapman 
and Hall. 
Manjunath, B.S. and Ma, W.Y.,1996. Texture features for 
browsing and retrieval of image data. Pattern Analysis and 
Machine Intelligence, IEEE Transactions on, 18(8): 837-842. 
Ozertem, U., Erdogmus, D. and Jenssen, R.,2008. Mean shift 
spectral clustering. Pattern Recognition, 41(6): 1924-1938. 
1220