115.

[Cover] [Contents] [Index]

Page 200

for using triangular areas to calculate the fractal dimension is shown in Figure 5.9.

The algorithms shown in Figures 5.5, 5.7 and 5.9 are used to generate unique measures of the fractal dimension for the entire image. However, for the purposes of image classification, we require an image showing a value of texture measured for the area surrounding each pixel, or at least for a number of subregions, rather than single overall fractal measurement. To generate a texture image using the fractal dimension approach, one can first divide the image into overlapped subimages. The fractal dimension can be derived separately for each subimage using one of the above algorithms, and the resulting fractal dimension is then assigned to the central pixel of the subimage (Figure 5.10a), thus building up a texture image. The pixel values in this image may then be mapped on to the interval [0, 255]. Note that the size of each subimage should not be too small as a small subimage will limit the number of window sizes used in the fractal dimension estimation procedure, and the result will naturally contain more bias. Conversely, if the subimage is too large it will exclude too many boundary pixels or overlap regions with different textures.

Figure 5.10b shows an HH-polarisation SAR image acquired by the Shuttle Imaging Radar C (SIR-C). The image is 256×256 pixels in size and shows part of the Red Sea Hills in Sudan. This area contains several lithological types, mainly ‘volcano-sedimentary’, ‘intrusives’, and ‘granitic batholith’, which demonstrate different types of texture patterns (Mather et al., 1998). The ‘intrusives’ rock type occupies the area around the lower-left corner of Figure 5.10b, and it is relatively smooth in appearance.

Figure 5.9 Estimation of fractal dimension using triangular area method.

[Cover] [Contents] [Index]