ÄS mW: S 
64X64 128X128 256X256 512X512 1024X1024 
This segmentation algorithm has the obvious parallel functional 
characteristics. 
Implementation of the algorithm in four PC nodes which consist 
of parallel programming environment is achieved, and the 
different nodes of computing time and computational speedup 
are analysis. There are specific terms of time and speed up as 
shown in fig 1. 
Fig 1 the implementation time of different size of image 
applying growth algorithm 
The algorithm is to use a series of seed through "data 
parallelism" to get the first division of the image. Then based on 
these parameters such as the divided region area, the average 
gray-scale by using "functional parallel", the image 
segmentation is completed. By analyzing figure 1, it is 
indicated that the greater image data, the greater the time 
speedup. Practice shows that the algorithm has good parallelism. 
5. CONCLUTION AND DISCUSSIONS 
Faced with the remote sensing data growing geometrically, it is 
an important topic to research efficient and fast image 
segmentation by regional growth. This paper summarizes the 
traditional regional growth segmentation algorithm. And the 
algorithm which suit for "data parallelism" and was raised 
based on grid environment. The practical result proves that the 
algorithm is suitable. The algorithm both in theory and in 
practice provides useful guidance and reference for image 
segmentation. 
Although growth image segmentation by regional has been a 
great deal of research results, but there are still some problems 
of parallel algorithm to be solved. 
(1) Seed selection and the effective growth criteria is still a hot 
and focus research spot. 
(2) Due to the application of this algorithm is still confined in 
small and single-band images, it is very necessary to study 
feasible and efficient parallel segmentation algorithm for the 
massive or multi-band remote sensing images. 
(3) Using "data parallelism" which Fully consider all the data- 
band and "functional parallel" which analysis the correlation 
among bands, the accuracy of the segmentation is expected to 
improve. 
REFERENCE 
[1] Zhang Y.J., 2001. Image Segmentation. Science Press, 
Beijing, pp67-70. 
[2] Yu Z.H., Chen Y., Liu P., 2003. Grid Computing. QingHua 
Press, Beijing, pp6-12. 
[3] Michael J. Quinn, 2004. Parallel Programming in C with 
MPI and OpenMP. QingHua Press, Beijing, pp 102-145. 
[4] Smith, J., 1987a. Close range photogrammetry for analyzing 
distressed trees. Photogrammetria, 42(1), pp. 47-56. 
[5] Nico, Giovanni; Fusco, Luigi; Linford, Julian, 2002. Grid 
technology for the storage and processing of remote sensing 
data: Description of an application. Proceedings of SPIE - The 
International Society for Optical Engineering, v 4881, pp. 677- 
685. 
[6] JaeHo Jeon, Hyung-Sun Kim, GeonYoung Choi and 
HyunWook Park, 2000. KAIST image computing system 
(KICS): A parallel architecture for real-time multimedia data 
processing. Journal of Systems Architecture, 46(15), pp. 1403- 
1418. 
[7] S. Boussakta, 1999. A novel method for parallel image 
processing applications. Journal of Systems Architecture, 
45(10), pp. 825-839. 
[8] Zhou H.F., Jiang Y.H., Yang X.J., 2002. Research on Serial 
and Parallel Strategies of Watershed Transform. JOURNAL OF 
NATIONAL UNIVERSITY OF DEFENSE TECHNOLOGY, 
24(6), pp. 71-76. 
[9] Zhou H.F., Liu G.M., Zheng M.L., 2004. A Research on 
Serial and Parallel Strategies of the Automatic Image 
Registration for Remote Sensing. JOURNAL OF NATIONAL 
UNIVERSITY OF DEFENSE TECHNOLOGY, 26(2), pp. 57- 
61. 
[10] A. Bevilacqua and E. Loli Piccolomini. Parallel image 
restoration on parallel and distributed computers. Parallel 
Computing, Volume 26, Issue 4, March 2000, Pages 495-506. 
TEXT 
Key Wor 
ABSTRA 
Higher sp 
same firm 
texture fe; 
different c 
seen that t 
As the ] 
technolog; 
reached 0. 
higher acc 
informatic 
more obvi 
As the re: 
easier to 
expert, bi 
difficult ti 
The reaso 
spectral ir 
textural ir 
order to in 
is necessi 
processing 
segmentai 
features ar 
Many text 
filed. Te» 
method, £ 
process m 
methods h 
processing 
simplified 
Texture ar 
applied 
Classificat 
proposed 
extraction 
The key p 
efficiently 
Gray level 
become oi 
features. 1 
correlation