ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001 
146 
vertical components of each shifting vector 
respectively. To each of the clusters in Px-Py space, 
we find the cluster center, or median vector, with 
which every member of the cluster should go along. 
4. Experimental Results 
To test our algorithms, we have simulated the Nojima fault, 
which accompanied the great Hanshin-Awaji earthquake of 
1995 in Japan. The model has been constructed with a 
rectified image which has been oriented by 
photogrammetric approach. Firstly, the shifting amount in 
the real world coordinates is specified. Then the shifting 
amount in image coordinates is calculated for the 
designated fault area. Lastly, move one side of the 
simulated fault along its direction by the calculated amount 
- in Nojima’s case both 1m in planar and horizontal as 
shown in Fig.3 (b). This model cannot ensure a uniform 
shift amount within the shifted area. Yet in this case the 
differences are negligible since the simulated area is very 
small when compared with the original image. 
With the pair images Fig.3(a);before the fault, and (b);after 
the fault, we applied the bi-cluster approach, to find out the 
fault boundary successfully as shown in Fig.3(c). 
5. Conclusion 
By introducing the new consensus operations into our 
nonlinear mapping technique, we were able to detect the 
simulated fault from a pair of aerial images. Since this 
technique uses a self-organizing nonlinear mapping based 
on the principle of coincidence enhancement, differences 
in the image distortion and non-equalized quality between 
two images can be automatically compensated for [3]. 
This work was partly supported by JST, Japan. 
References: 
[1] Y. Kosugi, P. Tchimev, M. Fukunishi, S. Kakumoto and 
T. Doihara: An Adaptive Nonlinear Mapping Technique for 
Extracting Geographical Changes; Proc. GIS2000 / 
CD-ROM; 4pages (2000) 
[2] M. Matsuoka and F. Yamazaki: Use of Interferometric 
Satellite SAR for Earthquake Damage Detection, Proc. 6 th . 
Conf. Seismic Zonation (2000) 
[3] Y. Kosugi, M. Sase, H. Kuwatani et a!.: Neural Network 
Mapping for Nonlinear Stereotactic Normalization of Brain 
MRI mages, J. Comp. Ass. Tomogra., Vol.17, No.3, 
445-460 (1993) 
(a) before earthquake (b)after earthquake (c) detected dislocation 
Fig.3 Automatically detected dislocation from a set of aerial photo images simulating the Nojima fault related to the 
great Hanshin-Awaji earthquake.