Color information and variances of edge intensity were analyzed from images of selected area, because the edges of middle intensity (32-90) existed in not only collapsed buildings but also in nondamaged buildings. Variances in the edge intensity were examined for the are of 7x7 pixels and were allocated a 1 byte value. The relationship between the variance value of edge intensity and the cumulative percentage of relative frequency is shown in Figure 5. for the collapsed buildings ( C-1,2,3), it is found that the variance values of edge intensity are almost all (90%) distributed in the range of 0-15.

(a) Distribution of relative frequency of hue for the selected buildings



(b) Distribution of relative frequency of intensity for the selected buildings



(c) Distribution of relative frequency of saturation for the selected buildings

Figure 6 Characteristics of color information for selected buildings



Figure 7 Edge detection for the selected areas
Color information such as hue, brightness and saturation as calculated using the RGB values for selected areas, and the relative frequencies of each area were calculated as shown in Figure 6. the relative frequencies of the hue of collapsed buildings are distributed in the range of 127.5-167.5 degrees ( this color range is from red to yellow ) . these values are justified by the fact that the soil under the roofing of the collapsed buildings could be observed. As for brightness and saturation, a difference was not confirmed between the values for collapsed buildings and for nondamaged buildings.

Figure 8 Distribution of damaged buildings extracted by image processing



Figure 9 Distribution of severely damaged buildings based on ground surveys (AIJ and CPIJ, 1995; BRI, 1996) and collapsed buildings based on visual extraction from aerial HDTV images
5. Extraction of Damaged Buildings Distribution by Image Processing
The edge elements which existed only in the area of collapsed buildings were extracted using conditions such as the edge intensity value in the range of 32-90, the variance values of edge intensity in the range of 0-15 and the hue value in the range of 127.5-167. the edge elements extracted by this procedure for the selected areas of collapse buildings. The damaged buildings distribution in this studied area was extracted using this procedure and is shown in Figure 8. also, severely damaged buildings based on ground surveys, and collapsed buildings based on visual extraction from aerial HDTV images, are shown in Figure 9. the extracted damage distribution data from aerial HDTV images agreed with damage survey results.

6.Conclusions
Image characteristics of damaged buildings were examined using image processing of aerial HDTV images taken after the 1995 Kobe earthquake. Furthermore, the distribution of damaged buildings was extracted on the basis of these characteristics. Characteristics of the damage to wooden buildings were defined on the basis if hue, edge intensity and edge intensity variance and by using the threshold value of these parameters. As for the detection of damaged-building distribution by using these characteristics, although the extraction accuracy was not particularly high, many edges extracted from the images of collapsed buildings and part of the severely damaged buildings.

Acknowledgment
The ground survey data used in this study were provided by the Architectural institute of Japan, the city planning Institute of Japan , and the Building Research Institute.

References

• Aoki, H., Matsuoka, M. and Yamazaki, F. (1998). Characteristics of satellite SAR images in the damaged area due to the hyogoken-nanbu earthquake, 19th Asian conf. Of remote sensing , pp.C7/1-6.
• Architectural institute of Japan and city planning Institute of Japan, (1995). Report of emergency damaged survey of the 1995 Hyogoken -Nanbu Earthquake.
• Building Research Institute, (1996). Final report of damage survey of the 1995 Hyogoken-Nanbu Earthquake ( in Japanese )
• Hasegawa, h., Yamazaki, F., Matsuoka, M., and Sekimoto, I. (1999). Extraction of Building Damage due to the 1995 Hyogoken-Nanbu Earthquake from Aerial HDTV Images, Processing of Asian-Pacific Symposium on structural Reliability and its Application NCREE, pp. 280-288.
• Hasegawa, h., Yamazaki, F., Matsuoka, M., and Sekimoto, I. (2000). Extraction of Building Damage due to earthquake using aerial television images, Proceedings of 12the world Conference on Earthquake engineering ( in press )
• Matsuoka, M. and Yamazaki, F. (1998). Identification of Damaged Areas due to the 1995 Hyogoken -Nanbu Earthquake using Satellite Optical Images, 19th Asian Conf. of Remote sensing, pp. Q9/1-6.