Geoscience Reference
In-Depth Information
High pass fi lter was applied to satellite images in frequency domain to delineate
the border of linear objects like roads and buildings precisely. Filters were applied to
the low frequencies which are around the center for both pre- and post-earthquake
data. After the fi ltering, IFFT was applied and edited Fourier images were converted
back into the spatial domain.
After the visual interpretation of IFFT images, a difference image was generated
by subtracting post-IFFT-image and pre-IFFT-image. Level slicing method with 10
slices was conducted to identify the damaged and non-damaged areas. Different num-
bers of levels were tried to fi nd out the optimum number of slices and the analyses
shows that having a slice number higher than 10 did not contribute signifi cant infor-
mation since only a few number of pixels was assigned to a slice. The histogram of
the difference image was investigated to see the general distribution of data and to
determine a threshold value for damaged and non-damaged regions. Further analyses
were conducted with different threshold values to fi nd out the most appropriate value
for the study to identify the changes. Standard deviation (σ) and mean (μ) values
obtained from the difference image were used for the analysis. 3σ, 2.5σ, 2σ, 1.8σ and
1.6σ, 1.5σ, and1.4σ were tried and 1.4σ was found as the best threshold value to de-
termine changes. Using this threshold value slices including data between μ-1.4σ and
μ+1.4σ were assigned as non-damaged areas whereas slices outside this range were
assigned as damaged areas. Figure 5 shows the result of level sliced-difference image
and blocks overlaid on this image with parcel boundaries.
Figure 5. Blocks overlaid on the difference image.
 
Search WWH ::




Custom Search